racial diversity, electoral preferences, and the supply of
TRANSCRIPT
Racial Diversity, Electoral Preferences, and the Supply of Policy:
The Great Migration and Civil Rights∗
Alvaro Calderon† Vasiliki Fouka‡ Marco Tabellini§
October 2020
Abstract
Between 1940 and 1970, during the second Great Migration, more than 4 million AfricanAmericans moved from the South to the North of the United States. The Great Migrationcoincided with the rise of the civil rights movement, whose success marked a key turningpoint in the history of race relations in the US. In this paper, we show that these twoevents are causally linked. Predicting Black in-migration with a version of the shift-shareinstrument, we find that the Great Migration increased support for the Democratic Partyand encouraged pro-civil rights activism, among both Black and (some) white voters. Wealso show that Black in-migration induced northern Congress members to more activelypromote civil rights legislation. However, these average effects mask a steep rise in polar-ization. While the 1940s saw the replacement of moderate Republicans with increasinglyliberal Democrats, the 1950s were characterized by a right-wing shift within the GOP. Over-all, our findings suggest that, under certain conditions, cross-race coalitions can emerge, butalso that higher support for racial equality may coincide with higher political polarization.
Keywords: Race, diversity, civil rights, Great MigrationJEL Classification: D72, J15, N92
∗We are extremely grateful to Nicola Gennaioli and Jim Snyder for several insightful suggestions and formany constructive conversations. We also thank Alberto Alesina, Leah Boustan, Melissa Dell, Ryan Enos, JeffFrieden, Luigi Guiso, Petra Moser, Markus Nagler, Gerard Padro i Miquel, John Parman, Torsten Persson,Vincent Pons, Evan Taylor, Gaspare Tortorici, Nico Voigtlaender, Matt Weinzierl, Gavin Wright and seminarparticipants at Berkeley, Bologna, EIEF, Harvard, LMU, LSE, Northwestern Economic History Lunch, PSEMigration Seminar, Rochester, Stanford, UPF, Virtual Seminars in Economic History, Warwick, the Atlanta2019 EHA Annual Meetings, at the Galatina Summer Meetings, the Yale Politics and History Conference,and the 2020 Virtual World Congress of the Econometric Society for useful comments. We are grateful toEric Schickler and Kathryn Pearson for sharing with us data on signatures on discharge petitions, and toJames Gregory for sharing datasets on NAACP presence and CORE non-violent demonstrations. Silvia Farina,Ludovica Mosillo, Monia Tomasella, Francesca Bramucci, Pier Paolo Creanza, Martina Cuneo, Sarah O’Brein,Gisela Salim Peyer, and Arjun Shah provided excellent research assistance. All remaining errors are ours.
†Stanford University. Email: [email protected]‡Stanford University, Department of Political Science. Email: [email protected]§Harvard Business School and CEPR. Email: [email protected]
1 Introduction
Racial inequality is a pervasive feature of US society, encompassing most of its domains – from
earnings to employment opportunities, from intergenerational mobility to incarceration rates.1
One of the potential causes of the racial gap and of its unwavering persistence is the lack of
political empowerment of Black Americans who, for a major part of American history, have
been denied even the most fundamental civil right in a democracy, namely the right to vote
(Aneja and Avenancio-Leon, 2019; Keyssar, 2009).
Black political oppression was particularly strong in the US South. Writing in 1944, Swedish
economist and Nobel Prize winner Gunnar Myrdal argued that migrating outside the region
represented the most effective strategy for Black Americans to achieve racial equality and
finally gain political rights (Myrdal, 1944). According to Myrdal, “[t]he average Northerner
does not understand the reality and the effects of such [Southern] discriminations”, and “[t]o
get publicity is of the highest strategic importance to [Blacks]”. Around this time, African
Americans started to move from the South to the North and West of the United States, hoping
to reach a “Promised Land” and leave behind them the system of disenfranchisement, violence,
and discrimination perpetuated by the infamous Jim Crow laws (Boustan, 2016; Wilkerson,
2011). Eventually, more than 4 million Black Americans migrated between 1940 and 1970 in
what is known as the Second Great Migration (henceforth, Great Migration).
The Great Migration temporally coincided with the development and eventual success of
the civil rights movement – a turning point in the history of race relations in the United States,
which culminated in the passage of the Civil and Voting Rights Acts of 1964 and 1965. Given
the resistance of southern politicians to extend the franchise to Black Americans, northern
legislators and grassroots organizations located in the North such as the National Association
for the Advancement of Colored People (NAACP) and the Congress of Racial Equality (CORE)
played a key role in the process of enfranchisement (Lawson, 1976). Was Myrdal right? Did
northward migration allow African Americans to gain political power?
In this paper, we study this question, analyzing the political effects of Black in-migration
to the US North and West between 1940 and 1970. Our analysis proceeds in two steps.
First, we examine how the Great Migration affected demand for civil rights and racial equality
among northern voters. We measure support for civil rights in several ways, but use as main
proxies the Democratic vote share in Congressional elections and the frequency of non-violent
pro-civil rights demonstrations organized by inter-racial grassroots organizations such as the
CORE. Even though the Democratic Party was openly segregationist and stubbornly defended
white supremacy in the South until the early 1960s (Kuziemko and Washington, 2018; Lawson,
1976), by the end of the 1930s in the North and West it had unambiguously become the party
defending Blacks’ interests and pushing for racial equality (Schickler, 2016; Wasow, 2020).2
Second, we analyze the effects of Black in-migration on the ideology and behavior of members
1See, among others, recent works by Bayer and Charles (2018) and Chetty et al. (2020). Previous, importantcontributions on this topic include Smith and Welch (1989) and Neal and Johnson (1996). See also the reviewin Altonji and Blank (1999).
2Below, we corroborate this idea, and provide evidence consistent with the existing literature (Feinstein andSchickler, 2008; Schickler, 2016). On party realignment during this historical period, see also recent work byCaughey et al. (2020).
1
of the House on race-related issues.
The political effects of the Great Migration are far from obvious. On the one hand, recent
work in economics has documented that the Great Migration had substantial, negative effects
on African Americans in the long run. Black in-migration to northern cities increased racial
residential segregation, as white residents fled urban areas for the suburbs (Boustan, 2010). In
turn, whites’ residential choices, coupled with changes in the allocation of local public goods
away from education and towards policing, drastically limited opportunities for economic and
social mobility of African Americans (Derenoncourt, 2018). Racial residential segregation and
lower economic opportunities may have been accompanied by whites’ political backlash and a
reduction in Blacks’ political efficacy.
On the other hand, the Great Migration might have promoted Blacks Americans’ political
empowerment for at least two reasons. First, around 1940, Blacks were de facto or de jure
prevented from voting in most southern states (Cascio and Washington, 2014), whereas no
restrictions to their political participation existed in the North. The inflow of Black voters
may have thus shifted (northern) politicians’ incentives to introduce civil rights legislation.
Second, Black arrivals may have moved the preferences of at least some white voters in a more
liberal direction. This might have happened either because the Great Migration increased
whites’ awareness of the conditions prevailing in the South, as envisioned by Myrdal (1944), or
because progressive parts of the Democratic coalition saw an opportunity to jointly promote
racial equality and economic goals by forming a cross-race alliance, as suggested by the political
science literature (Adams, 1966; Frymer and Grumbach, 2020; Schickler, 2016).
To study the political effects of the Great Migration we estimate stacked first difference
regressions, controlling for state time-varying (observable and unobservable) characteristics,
and allowing counties to be on differential trends depending on their initial Black share and
political conditions. We construct a version of the shift-share instrument (Boustan, 2010;
Card, 2001) that assigns Black outflows from each southern state to northern counties based
on pre-existing settlements of African Americans outside the South.
The shift-share instrument does not merely apportion more Black migrants to counties
with more African Americans in 1940, but rather, it combines two separate sources of varia-
tion. First, it exploits variation in the “mix” of Blacks born in different southern states and
living in different northern counties in 1940. Second, it leverages time-series variation in Black
emigration rates from different southern states for each decade between 1940 and 1970. Since
we always condition on the 1940 Black share of the population — interacted with period dum-
mies — the instrument only exploits variation in the composition of Black migrants across
southern states over time.
The validity of the instrument may be threatened if the characteristics of northern areas
– such as the fraction of the workforce in manufacturing or the urban share – that attracted
a different mix of southern born African Americans before 1940 had persistent, confounding
effects both on changes in racial attitudes and on migration patterns (Goldsmith-Pinkham
et al., 2020). We address this issue in two ways. First, we document that predicted Black
in-migration is not correlated with the pre-1940 change in political conditions across northern
counties. Second, we allow counties to be on differential trends by interacting year dummies
2
with several 1940 local characteristics, such as the Black, immigrant, and urban share of the
population, initial support for the Democratic Party, and the employment share in manufac-
turing. We also interact year dummies with the share of Blacks born in each southern state
to check that the variation behind the instrument is not disproportionately driven by specific
destination-origin combinations, which may also be spuriously correlated with the evolution of
political conditions in the North.
The instrument may also be invalid if shocks to northern counties both influenced the local
economic and political landscape and caused outmigration from southern states that already
had large enclaves in those counties before 1940 (Borusyak et al., 2018). To tackle this potential
concern, we first perform placebo checks to show that the instrument is uncorrelated with local
WWII spending (one of the key “pull factors” that attracted Blacks northward during the Great
Migration, e.g. Boustan, 2016) and New Deal relief programs. Second, we replicate the analysis
separately controlling for a measure of predicted labor demand, constructed by interacting the
1940 industry composition of northern counties with nation-level industry growth rates after
1940. Finally, as in Boustan (2010) and Derenoncourt (2018), we construct a modified version
of the instrument that exploits only variation in local push factors across southern counties to
predict Black outflows from the US South.3
Given the vast literature on “white flight” (Boustan, 2010; Shertzer and Walsh, 2019), we
verify that the Great Migration did not lead to white out-migration at the county level. We also
document that the composition of white residents did not change in response to Black inflows.
Notably, these results are not in contrast with previous work (Boustan, 2010; Derenoncourt,
2018). We provide evidence that county boundaries do not overlap with city-suburbs divides.
This, in turn, implies that changes in population triggered by Black inflows occurred within
(and not between) the jurisdictions that we consider in our analysis.
Turning to our main results, we find that Black in-migration had a strong and positive
impact on the Democratic vote share in Congressional elections. Our estimates imply that
one percentage point increase in the Black share raised the Democratic vote share by 1.6
percentage points, or 4% relative to the 1940 mean. This is a large effect: even under the
aggressive assumption that all Black migrants immediately voted for the Democratic Party
upon arrival, support for Democrats must have increased among northern residents because of
Black inflows. Complementing our electoral results, we find that Black arrivals increased both
the frequency of non-violent pro-civil rights demonstrations from CORE and the presence of
local NAACP chapters.
Consistent with the view that African Americans were quickly incorporated in the politi-
cal life of northern cities (Moon, 1948), we show that Black in-migration had a positive but
quantitatively small and imprecisely estimated impact on turnout. This indicates that Black
in-migration likely induced existing voters to switch away from the GOP. Since not all Black
residents were already voting for the Democratic Party in the early 1940s, some switchers were
probably African Americans. However, the magnitude of our estimates implies that some seg-
ments of the white electorate likely joined the Democratic voting bloc as well. Using a subset
3By exploiting southern county specific shocks, this strategy also assuages potential concerns over serialcorrelation in migration flows from the same location to the same destination (Jaeger et al., 2018).
3
of the data on pro-civil rights demonstrations, which reports the race of participants, we find
that the propensity to join pro-civil rights demonstrations increased not only among Black but
also among white individuals. This is consistent with our previous interpretation that some
whites became more supportive of racial equality because of Black arrivals.
To understand which segments of the white electorate became more supportive of civil
rights, we explore heterogeneity patterns in our results. Focusing on pro-civil rights demon-
strations, we exploit variation in county 1940 composition and historical characteristics.4 First,
we consider the possibility that the Great Migration raised support for civil rights among so-
cially progressive whites by increasing the salience of the “race problem” and activating their
latent demand for racial equality (Allport, 1954; Myrdal, 1944). In line with this view, we
show that pro-civil rights demonstrations were concentrated in counties with a lower history
of racial discrimination, and where the white population was younger and less likely to come
from the US South.
Second, we document that pro-civil rights demonstrations were more likely to occur in areas
with a higher share of whites employed in manufacturing and in the unskilled sector, where the
presence of the CIO was stronger, and where elections were more competitive. These places may
have offered fertile grounds for the formation of a liberal cross-race coalition along political and
economic lines, as discussed extensively in Schickler (2016). Labor unions may have supported
a cross-race coalition only, or especially, when labor markets were tight (Bailer, 1944). Indeed,
our results are driven by counties where labor demand, predicted using a Bartik-style approach,
was stronger.
We corroborate our heterogeneity analysis by providing suggestive evidence from historical
survey data from the American National Election Studies (ANES) and from Gallup. Estimating
state-level cross-sectional regressions, we find that, in the years preceding the Civil Rights Act
of 1964, white respondents living in states that received more Black Americans between 1940
and 1960 held significantly more favorable views on race relations, considered racial equality as
one of the most fundamental issues for the country, and were more likely both to identify with
and to vote for the Democratic Party.5 These results are driven by liberal and Democratic
respondents, and by members of labor unions.
Our findings may seem at odds with the literature on white flight and the detrimental
consequences that the latter had on Black migrants and their offspring in the long run (Boustan,
2010; Derenoncourt, 2018).6 However, Black political empowerment and white flight are not
necessarily in contrast with each other. For one, there is extensive evidence that the Great
Migration did economically benefit Black migrants (Baran et al., 2020; Boustan, 2016; Collins
and Wanamaker, 2014). In addition, white voters may have supported civil rights, while at
the same time moving from central cities to the suburbs (within the same county). From
the lens of a Tiebout (1956) framework, white voters may have expressed their preferences
regarding residential segregation and local public spending (schooling, in particular) with their
4Since data on respondents’ race is available only for a limited subset of events, we conduct this heterogeneityanalysis for all demonstrations, and not just on those involving white (and Black) participants.
5We are unable to conduct this analysis at the county level, as the ANES reports county identifiers only from1978 onwards.
6The Great Migration also increased racial disparities in incarceration rates (Eriksson, 2019; Muller, 2012),and worsened public finances in northern cities (Tabellini, 2018). See Collins (2020) for a thorough review.
4
feet, while instead using the ballot box to express their more general ideological preferences
about political racial equality. Supporting this conjecture, pro-civil rights demonstrations were
more likely to take place in counties where 1940 residential segregation was higher. That is,
when inter-group contact in the housing market was lower, whites were more supportive of
civil rights. Moreover, Black in-migration increased the number of school districts in counties
with higher 1940 levels of residential segregation. These patterns indicate that whites in more
segregated counties simultaneously showed higher support of civil rights and founded new local
jurisdictions to further exclude Blacks.
In the second part of the paper, we turn to the effects of the Great Migration on the ideology
and behavior of legislators representing non-southern congressional districts (CDs). Similar to
Autor et al. (2020), we construct a cross-walk that matches counties to CDs, and then develop
a procedure that assigns CD boundaries, which changed over time due to redistricting, to the
geography of a baseline, the 78th, Congress.7 We measure legislators’ ideology using the civil
rights scores constructed by Bateman et al. (2017), which are based on a legislator’s past voting
behavior on civil rights bills, and take more negative (resp. positive) values for more liberal
(resp. conservative) ideology. Mirroring our results in the electorate, over time, CDs that
received more African Americans were represented by legislators with a more liberal ideology
on racial issues over time. To more directly capture legislators’ behavior, we also show that
Black in-migration had a strong, positive effect on the probability of signing discharge petitions
aimed at promoting civil rights bills (Pearson and Schickler, 2009; Schickler, 2016).
Exploring the heterogeneity behind these average effects, we obtain two sets of results.
First, comparing legislators in CDs that remained Democratic throughout the period to leg-
islators representing CDs that always remained Republican, the ideology of the latter moved
significantly to the right on racial issues relative to that of the former. Second, as in Autor et al.
(2020) and in Tabellini (2020), we classify legislators in four categories – liberal Democrats,
moderate Democrats, moderate Republicans, and conservative Republicans – and show that
Black in-migration increased the probability of electing a liberal Democrat in the 1940s, but
a conservative Republican in the 1950s. Since quantitatively the first effect prevailed over the
second one, on average, legislators’ ideology moved to the left. However, when inspecting these
dynamics more carefully, polarization becomes evident.
This paper contributes to various strands of literature. First, it is related to the literature
on the civil rights movement. Several papers have studied the consequences of the Civil Rights
and the Voting Rights Acts (Aneja and Avenancio-Leon, 2019; Bernini et al., 2018; Cascio
and Washington, 2014; Cascio et al., 2010; Reber, 2011), while many others, building on the
seminal contribution by Carmines and Stimson (1989), have investigated the causes of the
southern “dealignment” (Besley et al., 2010; Kousser, 2010; Kuziemko and Washington, 2018;
Trende, 2012; Wright, 2013). We contribute to this literature by examining one of the causes of
the civil rights movement, and showing that the Great Migration likely played a key role in the
success of the latter. Our findings are also consistent with and complement Schickler (2016)
and Grant (2020) who, respectively, argue that the incorporation of African Americans into
the Democratic coalition after the New Deal and the rising pivotal role of African American
7The appendix describes this procedure in detail and performs a number of robustness checks.
5
voters at the national level due to the Great Migration were important mechanisms behind the
racial realignment of American politics.
Second, we complement the vast and growing literature on the political effects of migration
and the broader literature on inter-group relations. Several papers find that immigration and
a larger size of the minority group can lead to backlash among natives or majority members
(Arzheimer, 2009; Dustmann et al., 2019; Enos, 2016; Tabellini, 2020). We instead show that,
under certain conditions, inter-group contact can favor the formation of cross-race social or
political coalitions, raising demand for racial equality also among members of the majority
group.8 Several factors can explain the difference between our findings and those in the existing
literature. First, Black in-migration might have increased whites’ awareness of the conditions
prevailing in the South (Myrdal, 1944). Second, the civil rights legislation was, by and large,
about the South, and northern white voters would have been only indirectly – if at all –
affected at least before 1965. Third, labor unions had political and economic incentives to
incorporate African Americans in their rank and files (Adams, 1966; Bailer, 1944; Schickler,
2016), forging a shared working class identity and pursuing common goals, conditions that
contributed to positive inter-group contact (Allport, 1954). Finally, our average effects mask
substantial heterogeneity. On the voters’ side, support for civil rights increased more in more
segregated counties. On the side of legislators, the Great Migration substantially increased
polarization along party lines.
Our results also speak to the literature on the relationship between voters’ demand and
politicians’ behavior (Caughey and Warshaw, 2018; Jones and Walsh, 2018; Kroth et al., 2016;
Lott and Kenny, 1999; Mian et al., 2010; Miller, 2008). Closest to our paper, Cascio and
Washington (2014) have documented that the Voting Rights Act (VRA) shifted the distribution
of local spending across southern counties towards Blacks, once the latter became eligible to
vote. We expand on their findings by focusing on the US North rather than the South, and by
analyzing one of the potential causes, rather than consequences, of the VRA – i.e., the response
of northern politicians to the change in the characteristics, and thus in the demands, of their
constituency due to Black in-migration.
Finally, we complement the literature on the Great Migration, recently reviewed in Collins
(2020). Although several papers in economics have studied its effects on the residential decision
of whites, intergenerational mobility, immigrant assimilation, and public finance (Boustan,
2010; Shertzer and Walsh, 2019; Derenoncourt, 2018; Fouka et al., 2018; Tabellini, 2018), little
evidence exists on its political effects. Our paper seeks to fill this gap, focusing on the role that
the Great Migration played in the development and the success of the civil rights movement.
The paper proceeds as follows. Section 2 describes the historical background. Section 3
presents the data, and Section 4 lays out the empirical strategy. Section 5 studies the effects
of the Great Migration on demand for civil rights legislation, and verifies that Black inflows
were not associated with white flight at the county level. Section 6 investigates how Congress
members responded to Black in-migration. Section 7 concludes.
8Our findings are consistent with those in Lowe (2017), Rao (2019), and Steinmayr (2020) from India andAustria respectively. We complement them by providing evidence from the US and in an instance where groupboundaries are defined by race rather than by caste, income, or refugee status.
6
2 Historical Background
2.1 The Great Migration
Between 1940 and 1970, more than 4 million African Americans left the US South for northern
and western destinations. This unprecedented migration episode is usually referred to as the
(Second) Great Migration. From 1915 to 1930, the First Great Migration brought to the
North 1.5 million Blacks. However, the Second Great Migration – from now onwards the
Great Migration – was substantially larger in magnitude and had more profound implications
for American politics and race relations (Boustan, 2016). Most Black migrants moved to
urban centers in the Northeast and mid-West, but the Great Migration was a geographically
widespread phenomenon, which affected also the West and less urbanized areas outside the
South (Figure 1).9
Black migrants were pulled to the North and West by economic opportunities and pushed
out of the South by racial oppression, political disenfranchisement, and poor working conditions
(Boustan, 2016). On the one hand, the outbreak of WWII increased demand for labor in
northern and western factories, raising the potential gains from migration. Even after the
WWII-related labor demand shock was over, higher expectations of upward social and economic
mobility kept attracting African Americans to the North at least until the late 1960s. On the
other hand, widespread violence and disenfranchisement, together with a separate and unequal
school system, provided strong incentives for Blacks to leave the South (Feigenbaum et al.,
2020; Margo, 1991). Moreover, the mechanization of agricultural harvest in the 1940s and
1950s reduced demand for labor in the already depressed southern agricultural sector, further
increasing the pool of prospective migrants (Grove and Heinicke, 2003; Whatley, 1985).
Out-migration from the South was strongest during the 1940s, with a Black emigration
rate of almost 15%, but remained high until the late 1960s (Figure A.1). As a consequence of
this migration episode, during which the US South lost 40% of its 1940 Black population, the
racial profile of the United States changed dramatically. While only 25% of African Americans
were living outside the South in 1940, this figure had increased to more than 50% by 1970.
On average, the Black share of the population in northern and western cities moved from less
than 4% to more than 15% in just three decades. These numbers were an order of magnitude
higher for main hubs like Chicago, Detroit, or St. Louis, where the Black share moved from 8,
9, and 11% to 32, 43, and 41% respectively (Gibson and Jung, 2005).10
2.2 Black Migrants and Northern Politics
The demographic change induced by the Great Migration had the potential to alter the political
equilibrium, especially in industrial and urban centers. In the US South, Blacks were de facto
and de jure disenfranchised through the use of literacy tests, poll taxes, and grandfather clauses
(Cascio and Washington, 2014; Lawson, 1976). On the contrary, they could, and in fact did,
9When defining the US South, we follow the Census classification but, as in Boustan (2010), we excludeMaryland and Delaware – two states that received net black inflows during the Second Great Migration. SeeTable A.1 for the list of southern states considered in the paper.
10In rural counties, the Black share remained substantially lower and rarely exceeded 2 or 3%.
7
vote in the North (Moon, 1948). The literature on social movements has documented that
the enfranchisement of Black migrants increased both the organizational capacity of the civil
rights movement and pressures on local politicians (McAdam, 1982). During the First Great
Migration, both Democrats and Republicans had tried to include African Americans in their
voting bloc. However, since the New Deal, the Democratic Party had emerged as the party
better equipped to address Blacks’ demands outside the US South (Caughey et al., 2020;
Schickler, 2016).
Figure A.2 plots the share of northern Democrats (blue bars) and Republicans (red bars)
voting in favor of civil rights bills between Congresses 78 and 88 (see Table A.2 for the detailed
list of bills). Both in the 1940s and in the 1950s, Democrats in the North were more likely to
support civil rights bills.11 Using data from Pearson and Schickler (2009), Figure A.4 confirms
these patterns by focusing on signatures on pro-civil rights discharge petitions – another, more
direct, measure of legislators’ commitment to racial equality (Schickler, 2016).12 Non-southern
Democratic Congress members were at least 30 percentage points more likely than their Re-
publican counterparts to sign a discharge petition to promote civil rights legislation between
Congress 78 and Congress 82. The gap rose to more than 50 percentage points in the following
decade (Table A.3).
Recognizing the different party position on racial issues, Blacks in the North were signifi-
cantly more likely to support the Democratic Party. Existing evidence indicates that at least
70% of registered Black voters outside the South were voting Democratic already in 1936 –
a share that gradually increased over time (Bositis, 2012). Taking advantage of its initial
position, the Democratic Party successfully incorporated Black migrants in its voting bloc.
Democrats also benefited from the behavior of labor unions – the Congress of Industrial Or-
ganizations (CIO) in particular – that, since the late 1930s, started to actively incorporate
African Americans in their ranks.13
Abundant anecdotal evidence exists that labor unions openly endorsed civil rights and
backed African Americans in their fight for racial equality (Adams, 1966; Bailer, 1944). For
instance, J. Brophy, one of the CIO leaders, declared in 1944 that “behind every lynching is the
figure of the labor exploiter...who would deny labor its fundamental rights”. On a similar vein,
in 1942 Walter Reuther, a highly influential figure in the United Automobile Workers (UAW),
declared that “...[racial discrimination] must be put on top of the list with union security and
other major union demands” (Zieger, 2000). In line with these statements, evidence from the
Congressional Quarterly Almanac shows that, for the 42 cases in which the NAACP took a clear
position on a proposed piece of legislation between 1946 and 1955, the CIO openly took the
very same position in 38 cases, and never took a position conflicting with that of the NAACP
(Schickler, 2016). As a result, a class-based coalition, pushing for both racial and economic
liberalism, emerged. This gave additional leverage to Black activists and organizations such as
the NAACP and the CORE to exert pressure on northern Democrats to pursue the civil rights
11Figure A.3 documents that the pattern is reversed once the US South is included.12At a time when southern Democrats could block any proposed civil rights-related bill even before it reached
the floor of the House, discharge petitions were filed by northern legislators to circumvent congressional com-mittees, and move bills to the floor for a vote (Beth et al., 2003). For more details see Section 3.
13Using survey data from Gallup, Farber et al. (2018) document that, while non-southern whites were signif-icantly more likely than Blacks to be union members in 1940, this pattern had been reversed by 1960.
8
agenda.
3 Data
Our study combines data from several sources. In this section, we review the variables consid-
ered in our analysis, and we refer the reader to Appendix D for a more detailed description of the
surveys used in the paper. To analyze legislators’ behavior, we construct a time-invariant cross-
walk from counties to CDs, fixing CD boundaries to the baseline Congress of 1944 (Congress
78). Appendix B describes this procedure in detail.
Black in-migration and demographic variables. Data on Black and total population
as well as on other demographic variables for non-southern counties come from the County
Databooks, from Haines et al. (2010), and from the 1940 full count Census of Population
(Ruggles et al., 2015). We also collected data on Black migration rates from Gardner and
Cohen (1992) and Bowles et al. (1990) for 1940-1950 and for 1950 to 1970 respectively.
Electoral outcomes. Since the New Deal, Democrats had become the pro-Black party
outside the US South (Caughey et al., 2020; Moon, 1948). Such racial realignment was more
likely to emerge in Congressional than in nation-wide Presidential elections (Schickler, 2016).
Motivated by these observations as well as by the evidence presented in Section 2.2, we focus
on the Democratic vote share and on turnout (defined as the share of votes cast in the election
over the total number of eligible voters in the county) in Congressional elections. Data are
taken from Clubb et al. (1990). Since Census data are available at the decennial level, and
because Congressional elections are held every two years, we focus on electoral returns for exact
Census years from 1940 to 1970.
Local support for civil rights. We obtain measures of local support for the civil rights
movement from two sources. First, we use the dataset assembled by Gregory and Hermida
(2019) combining a variety of sources that includes the number of non-violent demonstrations
organized between 1942 and 1970 by the CORE – an inter-racial group of students from the
University of Chicago that coordinated sit-ins and similar forms of civil disobedience mainly
across northern cities to protest against segregation in the South. Second, we collect data on
the presence of NAACP chapters from Gregory and Estrada (2019). In both cases, we match
the geographic coordinates of an event or of a NAACP chapter to the centroid of each county
in our sample.14
Whites’ attitudes. We collect data on whites’ racial attitudes and stance on civil rights
from two, nationally representative surveys: the ANES and Gallup. Both are cross-sectional
datasets that report individuals’ socioeconomic and demographic characteristics as well as
their political ideology. Starting from the mid to late 1950s, both surveys began to elicit
respondents’ views on racial equality and their support for civil rights. Even though neither
dataset includes a county or city identifier, they both report respondents’ state of residence,
allowing us to correlate the change in the Black share at the state level with attitudes of
white respondents interviewed between the late 1950s and the mid-1960s (when the CRA was
14Data on NAACP chapters are available only for the early 1940s and the early 1960s.
9
passed).15
Legislators’ ideology. We measure the ideology of northern legislators on civil rights
by using the scores constructed by Bateman et al. (2017). As for the commonly used DW
Nominate scores (Poole and Rosenthal, 1985), legislators are assigned a score that is a function
of their past voting behavior and takes more negative (resp. positive) values for more liberal
(resp. conservative) positions. We rely on the Bateman et al. (2017) scores for two reasons.
First, they are calculated by restricting attention solely to civil rights bills, as classified by
Katznelson and Lapinski (2006). Second, they allow the policy content to be Congress specific
and to vary over time.16
Signatures on discharge petitions. Exploiting the seniority system prevailing at the
time, southern Democrats were able to block most proposed civil rights-related bills even before
they reached the floor of the House (Schickler, 2016). To overcome this obstacle, northern
legislators could rely on discharge petitions to move a bill to the floor, provided that a petition
collected at least 218 signatures (Beth et al., 2003). Using data from Pearson and Schickler
(2009), we measure a legislator’s involvement with (and support for) civil rights with signatures
on discharge petitions concerning racial issues.
Table 1 presents summary statistics for the main variables considered in our analysis, re-
porting 1940 levels in Panel A and their (decadal) changes in Panel B. The Black share in the
average county was around 3.5% in 1940, and increased to almost 9% in 1970 (not shown).17
These average values, however, mask substantial heterogeneity. Figure A.5 plots the 1940 Black
share for the counties in our sample, and shows that, in 1940, Blacks living outside the South
were concentrated in the urban centers of the Northeast and the Midwest, in border states like
Missouri and Kansas, and in southern California and some areas of Arizona and New Mexico.
For example, in Cook County (IL), the Black share in 1940 was already as high as 8%, and
rose to 21.5% by 1970. Similarly, the Black share in Philadelphia County (PA) increased from
more than 12% in 1940 to almost 35% in 1970, whereas that in Alameda County (CA) rose
from 2 to 15% during the same period (Figure A.6).
Turning to our main outcomes, the 1940 Democratic vote share in Congressional elections
was on average 45.7%. Focusing on the 78th Congress, our baseline Congress year, civil rights
scores were on average negative (-0.87), indicating that northern legislators were relatively
liberal on racial issues already by 1940. The average decadal change in ideology scores was very
close to zero, even though this masks important differences both between parties and between
Congress periods (Bateman et al., 2017; Schickler, 2016). Finally, signatures on discharge
petitions were significantly more common in the 78th- 82nd than in the 83rd - 88th Congress
period (Tables A.4 and A.5), and their subjects changed markedly over time. While the poll
15The restricted use version of the ANES includes county of residence of respondents from 1978 onwards.More details about the survey questions used in our analysis are provided in Appendix D.
16Bateman et al. (2017) develop two main versions of the score – one that assumes that the ideal points oflegislators remain constant over time, and one that instead does not make such assumption. As shown below,our results are robust to using either of the two versions.
17We drop non-southern counties that had no African American population in 1940, for which the instrumentfor Black in-migration (described below) cannot be constructed. Results are robust to restricting attention tocounties with a 1940 Black population above different thresholds or to including also counties with no Blackresidents at baseline.
10
tax and FECP legislation were the most common topics during the 1940s, 5 of the 8 discharge
petitions filed between the 83rd and the 88th Congress concerned the CRA.
4 Empirical Strategy
4.1 Estimating Equation
Our empirical analysis is divided in two parts. First, we estimate the effects of the Great
Migration on demand for civil rights legislation; second, we analyze the response of northern
legislators to changes in the composition and preferences of their electorate. To be clear: we
do not attempt to isolate the impact of changes in voters’ demand, due to Black inflows,
on legislators’ behavior. In fact, both parties likely re-optimized their platforms strategically
because of the Great Migration, in turn influencing the actions of voters – both Blacks and
whites. Our goal is instead to estimate the “reduced form” effect of Black in-migration on
voters’ demand and politicians’ supply without taking a stance on how the two influenced each
other.
Starting from the demand side and stacking the data for the three decades between 1940
and 1970, we estimate
∆ycτ = δsτ + β∆Blcτ + γXcτ + ucτ (1)
where ∆ycτ is the change in the outcome of interest in county c during decade τ . When
focusing on electoral outcomes, ycτ refers to the Democratic vote share and turnout in Con-
gressional elections. When considering grassroots activism, ycτ is the probability of pro-civil
rights demonstrations organized by the CORE and the presence of local NAACP chapters. In
order to identify the effects for the average county, we weigh regressions by 1940 county popu-
lation, but, as shown in Appendix C, results are robust to estimating unweighted regressions.
Standard errors are clustered at the county level.
The key regressor of interest, ∆Blcτ , is the change in the Black share in county c during
period τ . δsτ includes interactions between period and state dummies, and Xcτ is a vector of
interactions between period dummies and 1940 county characteristics. Our preferred specifica-
tion includes the 1940 Black share and a dummy equal to one if the Democratic vote share was
higher than the Republican vote share in the 1940 Congressional elections, but in Appendix C
we add more interactions to probe the robustness of our results. Since equation (1) is taken
in stacked first differences and always controls for interactions between period and state dum-
mies, the coefficient of interest, β, is estimated from changes in the Black share within the
same county over time, as compared to other counties in the same state in a given period.
Turning to the ideology and the behavior of legislators, we estimate a version of equation
(1), with a few differences. First, the unit of analysis, c, is no longer the county but, instead,
the CD.18 Second, when considering ideology scores, we restrict attention to two – rather than
three – periods. This allows us to end our analysis with the Congress that passed the CRA
(Congress 88). Third, when focusing on signatures on discharge petitions, we are forced to
18As noted above, we rely on the time-invariant unit of analysis constructed in Appendix B to deal withredistricting. Regressions are weighed by CD population, and standard errors are clustered at the CD level.
11
estimate equation (1) only for the 78-82 Congress period, when a sufficient number of petitions
were filed both at the beginning and at the end of the decade.
4.2 Instrument for Changes in Black Population
The key empirical challenge for our analysis is that Black migrants might have sorted in places
that were already undergoing economic and political changes. To overcome these and similar
concerns, we predict Black inflows in northern area c during decade τ using a version of the
shift-share instrument commonly adopted in the immigration literature (Boustan, 2010; Card,
2001). The instrument predicts the change in the Black population in county c during decade
τ by interacting the share of Blacks born in southern state j and living in northern county c
in 1940 (relative to all Blacks born in state j living outside that state in 1940), shjc, with the
number of Blacks who left state j during period τ , Bljτ :
Zcτ =∑
j∈SouthshjcBljτ (2)
Since we are interested in the effects of changes in the Black share, we scale Zcτ by 1940 county
population.
As discussed in Boustan (2010) among others, Black settlements in the North were highly
persistent over time. At the turn of the twentieth century, as African Americans started to move
northwards, migration patterns were influenced by the newly constructed railroad network. For
instance, the presence of the Illinois Central, which was connecting several Mississippi counties
to Chicago and a number of southern railroads to northern hubs in Missouri and Illinois,
explains why Black migrants from Mississippi were disproportionately concentrated in Chicago
or St. Louis (Grossman, 1991). The stability of Black enclaves was further reinforced by the
process of chain migration during the First Great Migration (Collins and Wanamaker, 2015).
Figure A.7 plots the share of Blacks born in Alabama, Mississippi, and Texas living in selected
northern counties in 1940, documenting the wide variation in settlement patterns across both
destination and origin areas.
4.2.1 Identifying Assumptions and Instrument Validity
Several recent papers have discussed the potential threats to the validity of shift-share designs
(Adao et al., 2019; Borusyak et al., 2018; Goldsmith-Pinkham et al., 2020; Jaeger et al., 2018).
In our setting, one threat to identification is that the characteristics of counties that pulled
Black migrants from specific states before 1940 may also be correlated both with post-1940
(Black) migration patterns from the South and with changes in support for civil rights in
northern counties (Goldsmith-Pinkham et al., 2020). We deal with this concern by performing
two sets of robustness checks, which are described in detail in Appendix C. First, we show that
pre-period changes in the outcomes of interest are not correlated with post-1940 changes in
Black in-migration predicted by the instrument. Second, we augment our baseline specifica-
tion by interacting year dummies with several 1940 county characteristics, such as the Black
share, support for the Democratic Party, the urban share of the population, and the share of
12
employment in manufacturing.
Controlling for the interaction between the 1940 Black share and year dummies implies
that the instrument only exploits variation in the (southern state) composition of African
Americans’ enclaves across counties, holding constant the size of their Black populations. We
also replicate our analysis by interacting year dummies with the share of Blacks born in each
southern state living in northern and western counties in 1940, i.e. shjc in equation (2). This
exercise reduces the concern that our estimates may be sensitive to variation coming from the
initial shares of Blacks born in specific southern states and concentrated in selected northern
counties (Goldsmith-Pinkham et al., 2020).
A second threat to the validity of the instrument is that it may be spuriously correlated
with specific shocks hitting northern counties that both affected local economic or political
conditions and influenced emigration patterns across southern states over time (Borusyak et al.,
2018). We address this concern in different ways. First, we document that the instrument is
uncorrelated with either WWII spending or the generosity of New Deal relief programs across
counties. Second, similar to Sequeira et al. (2020) and Tabellini (2020), we replicate the
analysis by separately controlling for a measure of predicted labor demand, constructed by
interacting the 1940 industrial county composition with the national growth rate of different
industries between 1940 and 1970. Third, as in Boustan (2010) and Derenoncourt (2018),
we replace actual outmigration from the South with that estimated by exploiting only initial
conditions across southern counties. This strategy also deals with the potential concern of
serial correlation in migration flows from the same location to the same destination (Jaeger
et al., 2018).
5 Demand for Civil Rights Legislation
This section studies the effects of the Great Migration on demand for civil rights. Section 5.1
documents that Black inflows increased the Democratic vote share in Congressional elections
and encouraged grassroots activism in support of racial equality. Section 5.2 verifies that the
Great Migration was not associated either with white flight between counties or with changes
in the composition of white residents, and summarizes additional robustness checks, which are
described in detail in Appendix C. Section 5.3 examines our results in more detail, considering
also the potential change in whites’ attitudes and political preferences.
5.1 Main Results
5.1.1 Congressional Elections
We start by studying the effects of the Great Migration on the Democratic vote share in Con-
gressional elections, which we interpret as a proxy for voters’ demand for civil rights legislation.
Panel A of Table 2 estimates equation (1) with OLS in columns 1 to 3, and with 2SLS from
column 4 onwards. Column 1 only includes state by year fixed effects, while columns 2 and 3
add interactions between year dummies and, respectively, the 1940 Black share and an indica-
tor for Democratic incumbency in 1940. In all cases, the point estimate on the change in the
13
Black share is positive and statistically significant.
Turning to 2SLS, Panel C of Table 2 shows that the instrument is strongly correlated with
the corresponding endogenous regressor, and the KP F-stat for weak instruments is always
above conventional levels. In our preferred specification – which includes interactions between
year dummies and: i) state dummies; ii) the 1940 Black share; and iii) an indicator for
Democratic incumbency in 1940 – the first stage coefficient implies that one percentage point
increase in the predicted Black share raises the actual Black share by 1.15 percentage points
(column 6).
2SLS estimates confirm the OLS results, but are larger in magnitude, especially for our
preferred specification (column 6), and when estimating long difference regressions (column
7). Our estimates are quantitatively large: according to our preferred specification (column
6), one percentage point increase in the Black share raised the Democratic vote share by
1.65 percentage points, or 4% relative to the 1940 mean. For large recipient counties such
as Cook county (IL) or Wayne county (MI), where the Black share increased by more than
15 percentage points between 1940 and 1970, Black in-migration likely altered the political
landscape dramatically.
The difference between OLS and 2SLS estimates indicates that Black migrants selected
areas where support for the Republican Party was rising faster. This might have happened
because these counties were experiencing faster income growth.19 Another possibility, not in
contrast with the previous one, is that the IV identifies a local average treatment effect (LATE)
for counties that received more Black migrants because of family networks and not because
of economic conditions. If Black Americans moving to a specific location due to the presence
of networks were more politically engaged relative to “economic migrants”, this could explain
why OLS coefficients are smaller than 2SLS ones.20
Panel B of Table 2 estimates the impact of Black in-migration on turnout. Our preferred
2SLS specification (column 6) indicates that Black arrivals had a positive but small and im-
precisely estimated effect on turnout in Congressional elections. As for the Democratic vote
share, OLS coefficients are smaller – in fact, even negative in this case – than 2SLS ones. The
quantitatively small effect on turnout is in line with existing evidence that Black migrants
were quickly incorporated in the political life of northern and western counties (Moon, 1948;
Schickler, 2016).
Table A.6 (columns 1 to 3) replicates the analysis separately for each of the three decades,
focusing on the preferred specification (Table 2, column 6). The Great Migration had a very
strong effect on the Democratic vote share in the 1940-1950 (column 1) and, though to a lesser
extent, in the 1960-1970 (column 3) decades. Conversely, the point estimate becomes smaller in
magnitude and not statistically significant for the 1950s (column 2). Turnout follows a similar
pattern, with a higher point estimate in the 1940s and in the 1960s, but results are imprecise
and never statistically significant.
19Corroborating this idea, in our sample there is a negative and statistically significant relationship betweenthe change in the Democratic vote share and a number of proxies for economic growth, such as populationgrowth, population density, and industrial expansion.
20Intuitively, one would expect political participation to be increasing in the size of one’s own network. Fur-thermore, migrants attracted solely by economic forces may remain more isolated from local (Black) communitiesand their political demands.
14
One interpretation of these findings is that the 1940s saw the dawn of the civil rights move-
ment, which was partly spurred by the Double V Campaign organized by African American
activists during WWII (Qian and Tabellini, 2020). The 1960s culminated with the passage of
the CRA and of the VRA and, even though in the later period whites’ backlash erupted in
many northern and western cities (Collins and Margo, 2007; Reny and Newman, 2018), this
may have been partly offset by greater engagement among Blacks. The lack of significance
and the smaller magnitude of the coefficient for the 1950s is consistent with the idea that the
economic downturn at least temporarily halted the progress of race relations, and cooled off
whites’ support for racial equality (Sugrue, 2014).
As discussed in Schickler (2016), support for racial equality was stronger within the local
fringes of the Democratic Party. Moreover, when it came to national politics, African Amer-
icans remained more skeptical about the commitment of Democrats to the civil rights cause.
Replicating the preferred specification of Table 2 for Presidential elections, column 4 of Table
A.6 confirms this idea. For the Democratic vote share (Panel A), the coefficient on the change
in the Black share remains statistically significant and positive, but is one third smaller than
for Congressional elections. The point estimate on turnout (Panel B) is somewhat larger, but
not statistically significant.
The positive effect of the Great Migration on the Democratic vote share likely reflects a
combination of i) migrants’ direct political engagement, and ii) existing residents’ changes in
voting behavior and preferences. The fact that Black in-migration increased the Democratic
vote share by more than one for one (Table 2, column 6) points to the importance of changes in
the voting behavior of northern residents. As noted in Section 2.2, the best available estimates
indicate that around 70% of northern Black registered voters supported the Democratic Party
in 1940 (Bositis, 2012). Moreover, turnout among African Americans was lower than average
(around 70% in 1940). This suggests that the Great Migration both induced a fraction of Black
GOP supporters to switch to the Democratic Party and increased the political engagement of
northern Blacks who were previously not voting, drawing them into the Democratic voting
bloc. The co-movement of the point estimates for the Democratic vote share and turnout (see
columns 1 to 3 in Table A.6) further highlights the importance of Blacks’ political participation
for the rise of the civil rights movement in the North.
According to the estimates reported in Table 2 (column 6), we can reject the null hypothesis
that the coefficient on changes in the Black share is equal to 1 at the 5% level. This suggests
that, besides the mobilization of Black northerners and the direct behavior of Black migrants,
the Great Migration induced at least some white residents to change their voting behavior too.
We return to these points in Section 5.3 below, when exploring the heterogeneity of our results.
5.1.2 Pro-Civil Rights Demonstrations and NAACP Chapters
In Table 3, we focus on the frequency of non-violent demonstrations organized by the CORE in
support of civil rights. The structure of the table mirrors that of Table 2, reporting OLS and
2SLS estimates in columns 1 to 3 and 4 to 7 respectively, and presenting first stage coefficients
in Panel B. For brevity, we focus on our 2SLS preferred specification (column 6).
Black in-migration had a strong, positive effect on the probability of CORE demonstrations.
15
One percentage point increase in the Black share led to a 4.7 percentage point increase in the
likelihood of protests. The CORE was created in 1942, and the frequency of events in our
sample of counties between 1942 and 1944 (included) was 0.09. Our estimates thus imply that
one percentage point increase in the Black share raised CORE demonstrations by more than
50% relative to their pre-1945 values. Another way to gauge the magnitude of these estimates
is to consider that the average change in the probability of CORE-sponsored demonstrations
in our sample is 0.143. Thus, one percentage point increase in the Black share can explain
almost one third of the change in pro-civil rights demonstrations across non-southern counties
between 1940 and 1970.
Using detailed information on the cause and the target of the protest available in the CORE
dataset, we classified the pro-civil rights demonstrations in different categories. Figure A.8
plots the number of events in each of the top four categories – discrimination in access to
goods and services (e.g. restaurants or hotels), school segregation, residential segregation, and
police brutality – as a share of all demonstrations in our sample.21 Each bar in the figure also
indicates the share of events within each category that concerned national (dotted bar area)
and local (Black bar area) issues. Almost two thirds of the events concerned local issues – such
as boycotting a local taxi company for its discriminating hiring process in Seattle or protesting
against a white-only barbershop in Chicago – but there existed substantial heterogeneity across
categories. For instance, while more than 80% of the events organized to demand a reduction
in residential discrimination were focused on local issues, almost 40% of demonstrations in the
“access to goods” category were conducted on a more national platform.
Relying on this classification, Figure A.9 replicates the analysis of Table 3 (column 6) for
each category separately. The first four dots from the left report 2SLS coefficients when the
dependent variable is the change in the probability of demonstrations for each of the causes
reported in Figure A.8. The remaining two crosses on the right report results for the change
in the probability of local and national demonstrations respectively.22 Even though the point
estimate is always positive, it is statistically significant and quantitatively larger for protests
against discrimination in access to goods and services and against school segregation (first
and second dots from the left). The coefficient is also larger and more precisely estimated for
demonstrations with local, rather than national, targets – something to be expected, since the
CORE was operating through local branches.
In Table A.8, we turn to the 1940-1960 change in the probability that a county had a
NAACP chapter in place, presenting OLS and 2SLS estimates in columns 1-2 and 3-4 respec-
tively, and reporting first stage coefficients in Panel B.23 In the full sample of counties, Black
inflows did not have any statistically significant effect on the local presence of NAACP chapters
(column 3). However, the impact of Black in-migration becomes positive, statistically signif-
icant, and quantitatively relevant for counties that did not have a chapter in 1940 (column
4). The fact that we do not find any effect for counties that already had a chapter in place in
1940 is not surprising. In these places, Black inflows likely increased the number of members
21Since events were classified according to either the cause or the target of the demonstration, the categoriesin Figure A.8 do not add to one.
22Table A.7 presents the corresponding 2SLS coefficients.23As noted in Section 3, data on NAACP chapters are only available for 1940 (or earlier) and 1960.
16
of NAACP chapters – something that we are not able to measure in our data. Instead, in
counties where the NAACP was not present at baseline, Black in-migration probably created
a critical mass of activists that justified the opening of new local chapters.
5.2 Robustness Checks
5.2.1 Addressing White Flight
A potential concern with results presented thus far is that Black arrivals induced white residents
to move to another county. Under this scenario, our estimates would conflate the causal effect
of Black in-migration with the compositional change in the county electorate driven by white
flight. This issue may be particularly poignant in our setting given the findings in Boustan
(2010), who documents that Black inflows to central cities induced white residents to move to
the suburbs. We thus need to verify that, at the county level, the Great Migration did not
systematically lead to white out-migration.
We begin by replicating the analysis conducted above focusing on a larger geographic
unit, the CZ, which typically contained both central cities and the suburbs.24 Table A.9
replicates Table 2, documenting that the effects of the Great Migration on the Democratic
vote share remain statistically significant and become, if anything, larger in magnitude.25 In
Table A.10, we conduct a similar exercise for CORE demonstrations. The precision of our
estimates deteriorates, perhaps not surprisingly given the local nature of such events and the
lower number of observations. However, 2SLS coefficients remain quantitatively very close to
those reported in the county-level specification of Table 3. Tables A.9 and A.10 suggest that
our main results are unlikely to be driven by white out-migration systematically triggered by
Black in-migration.
Next, to more directly inspect the presence of white flight, we replicate the analysis con-
ducted in Boustan (2010) for the counties in our sample. We regress the decadal change in
white population against the corresponding change in Black population. We consider the num-
ber of white and Black residents both to make our analysis directly comparable to that in
Boustan (2010) and because this is the most appropriate specification to examine the migra-
tion response of northern residents (see also Peri and Sparber, 2011, and Shertzer and Walsh,
2019). We report 2SLS results in Panel A of Table 4, presenting the associated first stage in
Panel B.
We start from a parsimonious specification, which only includes interactions between state
and year dummies (column 1). Panel B verifies that the instrument is strong, and the F-
stat is well above conventional levels.26 Turning to Panel A, 2SLS coefficients are positive,
quantitatively small, and imprecisely estimated. In column 2, we include the same set of
controls as in our preferred specification (see Section 5.1 above). Also in this case, Black
24CZs have become the standard measure of “labor markets” in the US since at least Autor and Dorn (2013).They were originally developed by Tolbert and Sizer (1996) using commuting patterns to create clusters ofcounties characterized by strong commuting ties within CZs and weak commuting ties across CZs.
25Panels B and C report, respectively, results for turnout and the first stage.26The point estimate in Panel B indicates that one additional predicted Black migrant is associated with 2.5
more Black residents in the county. The magnitude of the coefficient is smaller than, but in line with, thatreported in Boustan (2010).
17
in-migration is associated with a small, positive, and imprecisely estimated effect on white
population.
The bottom rows of Table 4 report the average 1940 white population and the average
change in Black and white population during the period for the counties in our sample. The
coefficient in column 2 (Panel A) implies that 1,000 more Black residents in a county – or, half
of the average change in Black population over the period – were associated with around 300
more white residents. Considering that, on average, the 1940 white population was 71,000,
this represents a negligible change (0.4% relative to the baseline white population). Columns 3
and 4 show that results are robust to including only counties with baseline urban share of the
population above the sample median (0.356), and to interacting the 1940 urban share of the
population with year dummies. Results are also unchanged when estimating long-difference
regressions (Table A.11).
Two observations help reconcile our findings with those in Boustan (2010). First, Boustan
(2010) focuses on central city to suburb migration, fixing city boundaries to 1940, whereas
we consider counties. Second, the (historical) central city-suburb divide does not overlap
with county boundaries; hence, the reallocation of white population between central cities and
suburbs was likely absorbed within counties. Table A.12 provides evidence consistent with this
conjecture. Specifically, in columns 1 and 2, we restrict attention to the 117 counties that are
included in the MSAs considered in Boustan (2010), and replicate our previous analysis. Also
in this sample, the Great Migration had no effect on changes in white population. In columns
3 and 4, we instead focus on central cities, and define the dependent variable as the change in
white population living there. Now, as in Boustan (2010), Black in-migration becomes strongly
associated with white out-migration.27
This analysis is reassuring because indicates that, at the county level, Black in-migration
did not trigger white out-migration. However, one may still be concerned that the Great
Migration led to selective white departures, which altered the composition of white residents.
To address this possibility we proceed as follows. First, we collect data from the 5% sample
of the 1960 Census of Population and from the full count Census of 1940.28 Given the limited
sample size and geographic coverage of the 1960 Census, we aggregate the data to the CZ and
conduct the analysis at this level.29 Next, restricting attention to white men above the age
of 18 and not enrolled in school, we create the share of residents in this group who were: i)
high skilled; ii) employed in manufacturing; iii) in the labor force; iv) homeowners; and v)
above the age of 65. Finally, we estimate long difference regressions, where the 1940 to 1960
change in each of the variables above is regressed against the corresponding (instrumented)
change in the Black share, including our preferred set of controls. 2SLS and first stage results
are reported, respectively, in Panels A and B of Table A.15. The coefficient on the change in
the Black share is always imprecisely estimated, quantitatively small, and does not display any
consistent pattern across outcomes.
27Results are unchanged when estimating long difference regressions (Table A.13).28For 1950 and 1970, only a 1% sample is available, limiting substantially the geographic coverage of the
datasets.29Not all CZs spanning the counties in our sample can be identified in the 1960 Census. Table A.14 shows
that restricting attention to the sample of CZs that can be identified in the 1960 Census leaves our politicalresults unchanged.
18
Using the approach just described, we also show that Black inflows did not increase labor
market competition for white residents (Tables A.16 and A.17 ).30 This confirms existing
evidence that northern labor markets were highly segmented along racial lines, and African
Americans rarely – if at all – directly competed for jobs with whites (Boustan, 2009).
5.2.2 Summary of Additional Robustness Checks
The previous section showed that our findings are unlikely to be influenced by white flight. We
now provide a summary of the additional robustness checks we conducted and that are more
extensively discussed in Appendix C.
First, we verify that the instrument is uncorrelated with three potential pull factors (either
for contemporaneous or for pre-1940 Black settlements): WWII contracts, New Deal spending,
and the vote share of Franklin D. Roosevelt in the 1932 elections (Table C.1).
Second, to address concerns that 1940 Black settlements might be correlated with county-
specific characteristics that may have had a time varying effect on changes in political condi-
tions, we interact period dummies with several 1940 county characteristics, such as the urban
and the immigrant share, the employment share in manufacturing, the employment to popula-
tion ratio, and county population (Tables C.2 and C.3). Tables C.2 and C.3 also document that
results are unchanged when augmenting the baseline specification with a measure of predicted
industrialization constructed using a Bartik-style strategy that combines the 1940 industrial
composition of non-southern counties with national growth across industries. Moreover, to deal
with the possibility that variation behind the instrument may be disproportionately driven by
specific northern counties-southern states combinations (Goldsmith-Pinkham et al., 2020), we
replicate our results by interacting year dummies with the share of Blacks from each southern
state (Figures C.1, C.2, and C.3).
Next, we show that there is no correlation between pre-period changes in the outcomes
of interest and the instrumented change in the Black share (Tables C.4 and C.11). Finally,
we provide evidence that our results are: i) robust to excluding potential outliers, estimating
different specifications, and measuring electoral outcomes in different ways (Tables C.5, C.6,
C.7, and C.8); ii) unlikely to be driven by the simultaneous inflow of southern white migrants
(Tables C.5 and C.6); and iii) unchanged when constructing a version of the instrument that
only exploits variation in “push factors” across southern counties (Tables C.9 and C.10).31
5.3 Mechanisms
Section 5.1 shows that the Great Migration increased support for civil rights among northern
voters. In this section, we first provide different pieces of evidence that Black in-migration
increased support for civil rights among some segments of the white electorate (Section 5.3.1),
and explore two channels through which this might have happened (Section 5.3.2). Next, we
show that (within county) residential segregation may have increased support for civil rights by
30As before, we restrict attention to men of age 18 or more who were not in school. Since data on employment,occupation, or wages are separately available by race (and gender or age) only from micro-censuses, we focus onyears 1940 and 1960, and conduct the analysis at the CZ level.
31Appendix C also performs additional robustness checks on CD results presented in Section 6.
19
making it easier for whites to create independent school districts, thereby lowering the potential
for racial animosity (Section 5.3.3).
5.3.1 Black in-Migration and Whites’ Attitudes Towards Civil Rights
Bounds on whites’ voting behavior. As already noted in Section 5.1, the fact that changes
in the Black share increased the Democratic vote share by more than one for one points to the
importance of changes in northern voters’ behavior. While northern Black residents certainly
played a role, in order to explain our previous estimates, at least some white residents had to
switch to the Democratic Party too. In Figure A.10, we provide bounds on the number of white
voters who had to switch from the Republican to the Democratic Party in order to match the
2SLS coefficient estimated in column 6 of Table 2 for the average county in our sample, under
different assumptions on turnout and voting preferences of African Americans.32
The red square on top of the four bars represents the number of votes for the Democratic
Party that are implied by our coefficient following one percentage point increase in the Black
share for the average county in our sample. The light blue (resp. dark blue) parts of each bar
represent the number of votes from northern Black (resp. white) residents in 1940, while the
grey area refers to votes for the Democratic Party cast by Black migrants. The area between
the top of the bar and the red square represents the number of voters – Blacks or whites – that
would have to switch from the GOP to the Democratic Party in order to explain our results. If
Black turnout was equal to the country-level average (70%), and 70% of all Blacks (migrants
and northern residents) voted for the Democrats (Bositis, 2012), our estimates would not be
fully explained by the behavior of Black voters, and we would also need some white “switchers”
(third bar from the left).33 For Black voters to fully explain our results, we would need to make
more extreme assumptions, such as above-average turnout or near universal support for the
Democrats among Black migrants (first and second bar from the left).
Clearly, this exercise is not meant to compute the exact number of white and Black voters
switching to the Democratic Party for each new Black migrant. Our goal is instead to show
that, under reasonable assumptions, Black migrants’ behavior alone is not sufficient to explain
the increase in the Democratic vote share estimated above, and that at least some northern
voters – both Blacks and whites – started to vote for the Democrats. In what follows, we
provide additional evidence consistent with this idea.
Additional evidence from CORE demonstrations. Section 5.1.2 above documented
that counties receiving more African Americans experienced a larger increase in the frequency
of pro-civil rights demonstrations organized by the CORE. While the CORE was an inter-
racial organization founded by both Black and white students from the University of Chicago
(Gregory and Hermida, 2019), our previous analysis remained silent on whether white activists
32When performing this exercise, we fix turnout, assuming that the inflow of Black migrants can change thepreferences of existing voters but does not alter the number of northern residents (of either race) voting. Thisassumption is consistent with our previous results for turnout (Table 2, Panel B).
33Specifically, under this scenario and the one in the fourth bar from the left, where we assume that Blackturnout was equal to 50%, approximately 2 white voters would have to switch for every ten Black migrants toexplain the 2SLS coefficient estimated in Table 2.
20
joined the demonstrations. We now make progress on this specific point by exploiting the fact
that, for a subset of CORE events, we were able to identify the race of participants.
In column 7 of Table 3, we estimate our preferred 2SLS specification considering as de-
pendent variable the change in the probability of CORE demonstrations with both Black and
white participants. Notably, this represents a (very conservative) lower-bound for the proba-
bility that whites joined pro-civil rights demonstrations, since only for approximately 40% of
CORE events participants’ race was reported, and we define a protest as having white partici-
pants only when their presence was explicitly reported. Although the point estimate is half as
that for the baseline specification (column 6), it remains positive and statistically significant,
with a p-value of 0.058.
Evidence from historical survey data. We complement the previous results by exploiting
historical survey data from the ANES and Gallup. As noted above, neither dataset reports
the county, but only the state of residence of respondents, and questions on racial views are
available only from the end of the 1950s.34 Thus, we estimate cross-sectional regressions,
correlating whites’ racial attitudes and political preferences in surveys conducted in years close
to the CRA with the (instrumented) 1940-1960 change in the Black share in their state of
residence. Because of the cross-sectional nature of this exercise, we acknowledge that the
evidence should be interpreted as suggestive.
To limit concerns about omitted variable bias, we always include survey year and Census
region fixed effects, a set of 1940 state level characteristics (manufacturing share, urban share,
share of unionized workers, Black share, indicator for Democratic incumbency in Congressional
elections), and individual controls (gender, marital status, and fixed effects for both age and
education).35 We always restrict attention to white respondents living in non-southern states.
We start by presenting results from the ANES in Table 5, where the dependent variable
is a dummy equal to 1 if support for civil rights was considered by respondents as one of the
most important issues for the country in 1960 and 1964 (see Appendix D for more details). In
the same survey years, respondents were also asked whether they opposed school and housing
or working space integration. Combining these questions, columns 1 to 3 of Table 5 first verify
that considering civil rights as one of the most important problems is strongly, negatively
correlated with opposition to racial integration. We thus interpret the dependent variable in
Table 5 as a proxy of support for racial equality. Next, columns 4 and 5 turn to the relationship
between the 1940-1960 change in the Black share and the most important issue dummy for
civil rights, using OLS and 2SLS respectively.
2SLS estimates indicate that white respondents living in states that received more Blacks
34The first year for which ANES data are available at the county level is 1978 – well after the period consideredin our analysis. Although some of these questions were asked also after 1964, we refrain from using any post-CRA survey dataset because of the potential direct effect of the bill on whites’ racial attitudes (Kuziemko andWashington, 2018; Wheaton, 2020).
35Since party identification and union membership may be endogenous to Black inflows, we do not includethem in our baseline specification. However, adding these controls does not change any of our results. Gallup didnot ask questions about marital status in the survey waves for which we could collect data on racial attitudes.Thus, this control is omitted when focusing on Gallup (see also Appendix D for more details). Results are alsorobust to including further 1940 state level controls such as the immigrant share, the share of unskilled workers,and other socioeconomic or political variables.
21
between 1940 and 1960 were significantly more likely to consider civil rights as one of the
most important issues. This relationship is quantitatively large: according to the coefficient
reported in column 5, one percentage point increase in the Black share between 1940 and 1960
is associated with a 1.8 percentage points (or, 17%) higher probability of considering civil
rights one of the most important problems in the two ANES surveys asked before the CRA. In
column 6 of Table 5, we exploit the fact that the ANES asked respondents not only their state
of residence, but also their state of birth. We restrict attention to whites who, at the time of
the survey, lived in the same state where they were born. Reassuringly, despite the reduction
in sample size, the coefficient on the change in the Black share remains statistically significant,
positive, and similar to that of column 5.36
Next, we focus on the relationship between the Great Migration and whites’ political pref-
erences. Focusing on survey waves between 1956 and 1964 and estimating 2SLS regressions,
Table A.19 documents that white respondents living in states that received more Black mi-
grants between 1940 and 1960 were significantly more likely to vote for (column 1) and identify
with (column 4) the Democratic Party. This relationship becomes slightly stronger when re-
stricting attention to non-movers (columns 2 and 5) and when considering only 1964 (columns
3 and 6). The fact that the relationship between Black inflows and whites’ support for the
Democratic Party was higher in 1964 is consistent with the civil rights issue featuring more
prominently during the year that led to the passage of the CRA.
Finally, Table A.20 verifies that similar patterns hold when using data from Gallup in the
1963 and 1964 waves. White respondents living in states that received more Black migrants
were less likely to object to the idea of racial mixing in schools (column 1), more likely to state
that they would vote for a Black president were their party to nominate one (column 2), and
more supportive of the CRA (column 3). There is instead no relationship between the Great
Migration and respondents’ views on whether the process of racial integration was proceeding
at the right pace (column 4).37
5.3.2 Unpacking the Channels Behind Whites’ Support for Civil Rights
The previous section provided different pieces of evidence that the Great Migration increased
support for civil rights among some segments of the white electorate. This might have hap-
pened for at least two, non-mutually exclusive, reasons. First, as envisioned by Myrdal (1944),
exposure to Black migrants might have increased whites’ awareness of the brutal conditions
prevailing in the South, in turn fostering demand for more racial equality in the region. In
addition, as predicted by the “contact hypothesis” (Allport, 1954), inter-group contact might
have reduced negative stereotypes and prejudice held by whites, changing their attitudes to-
wards Blacks. Second, economic and political incentives might have favored the formation
of a class-based cross-race coalition between white and Black Americans. This possibility is
36Table A.18 confirms the lack of correlation between Black in-migration and whites’ mobility across statesby regressing a dummy equal to 1 if the respondent lived in a state other than her state of birth against the(instrumented) 1940-1960 change in the Black share in the state of residence. This holds for different subsets ofrespondents (see the bottom row of the table).
37The number of observations varies substantially across questions, since some of these were asked repeatedlyduring 1963 and 1964. This was particularly true for the question on the pace of racial integration (column 4).
22
discussed, among others, in Adams (1966), Schickler (2016), and Sugrue (2008), who argue
that the labor movement in northern and western urban areas saw the Great Migration as a
window of opportunity to strengthen its political clout.
Clearly, economic and social factors may have interacted. For instance, frequent contacts
in an environment that promoted a shared, class-based, identity and where Black and white
workers had common goals reduced the barriers traditionally in place between members of
the majority and of the minority group, favoring the formation of a racially diverse social
coalition.38 Once a liberal coalition on both economic and racial issues was formed, its members
were likely in a better position than before to organize effective political campaigns. In turn,
the formation of a pro-Democratic coalition, favored by Black arrivals to northern cities, might
have put the Democratic Party in an advantageous position relative to the Republican Party.
In what follows, we focus on pro-civil rights demonstrations as our main indicator for voters’
demand for racial equality. We exploit heterogeneity in the 1940 characteristics of counties in
order to understand which segments of the white electorate increased their support for civil
rights.39
Social and cultural forces. According to the information mechanism envisioned by Myrdal
(1944), the Great Migration should have led to a larger progressive shift in whites’ racial
attitudes in counties that were already more socially progressive. To test this idea, we estimate
our preferred 2SLS specification, splitting the sample above and below the median of different
proxies for progressive attitudes in the local electorate. Figure 2 plots the coefficient on the
change in the Black share for counties above (orange bars) and below (blue bars) the median
of each proxy for socially progressive attitudes. To ease the interpretation of results, we always
rescale the variables along which we perform the sample splits so that higher (resp. lower)
values refer to socially more (resp. less) progressive counties.40
First, we consider the discrimination index constructed in Qian and Tabellini (2020) using
historical data from a variety of sources, such as local presence of the KKK and the lynching
of Black Americans up to 1940. We find that the effects of the Great Migration were an order
of magnitude larger in counties with lower historical discrimination. The same pattern, though
less pronounced, is evident when splitting counties as belonging to states with (blue bars)
and without (orange bars) miscegenation laws (Dahis et al., 2020). Second, we find that pro-
civil rights demonstrations increased more in places with a larger 1940 immigrant population.
Higher immigrant share captures tolerance of the local population, as the foreign-born are more
likely to sort into areas that are open to minorities. This result is also in line with historical
accounts noting that minorities were more likely to join African Americans in their demand
for racial equality (Schickler, 2016; Sugrue, 2008).41
38Using recent data, Frymer and Grumbach (2020) find that white union members hold significantly moreliberal attitudes towards minorities in the United States.
39Due to the limited number of events for which we have information on the race of participants, we are forcedto perform the heterogeneity analysis using data on all pro-civil rights demonstrations.
40The corresponding estimates are reported in Table A.21. Given that the values of the F-stat are often belowconventional levels, results should be interpreted with caution.
41Exceptions may be the Irish and the Poles, who were more likely to compete for status and jobs with BlackAmericans, and who had been historically antagonist to the latter (Bailer, 1943; Sugrue, 2014). In unreported
23
Finally, we explore how results are mediated by the characteristics of the white electorate.
The last two sets of bars in Figure 2 split the sample according to the 1940 share of whites
who were: i) young (defined as individuals of age 35 or younger); and ii) born in the South.
The former proxy is motivated by the fact that young individuals are more open to social
change (Acemoglu et al., 2020; Inglehart, 2015; Miles, 2000; Parker et al., 2016). The latter
is motivated by the fact that southern born whites were less likely to support racial equality
(Bailer, 1944; Kuziemko and Washington, 2018). Even though coefficients are less precisely
estimated than before, they indicate that CORE demonstrations increased more in counties
where the white population was more progressive.
Political and economic forces. The second mechanism that may have increased whites’
support for civil rights is that the labor movement strategically incorporated African Americans
in order to increase its political strength. This, in turn, led to a cross-race coalition pushing
for both economic and racial liberalism (Schickler, 2016). As before, we exploit baseline cross-
county heterogeneity, estimating the effects of Black in-migration on pro-civil rights protests
separately for counties above and below different proxies for the presence and the strength
of organized labor, or for its incentives to incorporate African Americans. We plot 2SLS
coefficients on changes in the Black share in Figure 3, always defining the variables over which
we conduct the split so that higher (resp. lower) values refer to stronger (resp. weaker) presence
of, or incentives for, unions to support the civil rights movement.42
The first set of bars from the left shows that pro-civil rights demonstrations were more likely
to occur in counties where political competition – defined as one minus the absolute value of the
margin of victory in 1940 Congressional elections – was higher. Instead, in counties with less
competitive elections, the Great Migration did not have any effect on pro-civil rights protests.
This finding is consistent with labor unions and the Democratic Party having stronger incentives
to coordinate events where the Black vote was more valuable. Precisely in these areas, a better
organized political machine could have made a difference in attracting and mobilizing pivotal
voters (Cantoni and Pons, 2020; Pons, 2018). Clearly, this argument is not confined to the
role of labor unions, but is true of organizational incentives for the civil rights movement more
broadly (McAdam, 1982).
Next, we turn to variables more directly related to the presence of labor unions. The second
and third set of bars from the left split the sample above and below the median of the share of
white men employed in manufacturing and in the unskilled sector, respectively. The surge in
civil rights demonstrations was concentrated in counties with a higher share of white workers in
the two sectors where unions were most widespread (Bailer, 1944; Farber et al., 2018). On the
contrary, Black in-migration had a negative, quantitatively small, and imprecisely estimated
impact on support for civil rights in counties with a low share of white men employed in
manufacturing or who were unskilled.
Abundant historical evidence shows that, among labor unions, the CIO was the most en-
results, we examined heterogeneity depending on the local presence of Irish and Pole first or second generationimmigrants, but we did not find any statistically or economically significant difference for the effects of the GreatMigration.
42We report the corresponding estimates in Table A.22.
24
gaged one with civil rights, and often worked side by side with inter-racial organizations like
the CORE and with NAACP leaders. Thus, one would expect Black in-migration to have a
stronger effect in places with a higher share of unionized workers who were members of the CIO.
Since CIO membership data are not available at the county level, we use the dataset assembled
by Troy (1957) to construct the share of unionized workers affiliated with the CIO in 1939 in
each state. Splitting the sample in counties belonging to states with CIO unionization rates
above and below the median, we find that the effects of the Great Migration were stronger,
although not statistically different, in areas where the CIO was relatively more important.
The last set of bars of Figure 3 tests a corollary of the “economic incentives” mechanism:
labor unions, and northern white workers more generally, should have been more willing to
accept Blacks and support racial equality when labor markets were tighter and economic op-
portunities more abundant. As conjectured by Allport (1954) and Blalock (1967), inter-group
contact is more likely to lead to cooperation when it happens in contexts with no competition
over scarce resources.43 Consistent with this prediction, the last bars of Figure 3 split the
sample in counties with predicted economic growth above and below the median, and show
that Black in-migration was associated with more pro-civil rights demonstrations only in the
former.44 These findings are consistent with a number of anecdotal accounts, such as Bailer
(1943) and Sugrue (2014), who point out that whites’ backlash was more likely to emerge dur-
ing economic downturns. They are also in line with the electoral results presented in Table A.6
above, which documented that the Great Migration had no effect on the Democratic vote share
in the 1950s – a decade characterized by slack labor markets and economic downturns.
We conclude this section by providing two more pieces of evidence on the strategic incentives
of labor unions using the ANES data. First, we investigate the heterogeneity behind results
presented in column 5 of Table 5, which showed that the Great Migration was positively
correlated with the probability that white respondents considered the civil rights issue as the
most important problem for the country. Figure A.11 documents that these patterns are
significantly stronger for union members (second, Black bar) and for Democrats (fourth, blue
bar).45 If anything, for Republicans we observe a negative, albeit not statistically significant,
relationship between state level increases in the Black share and support for civil rights. Since
union status and partisanship may be endogenous to the Great Migration, results in Figure A.11
should be viewed as merely suggestive. However, they paint a picture coherent with our
previous discussion.
Second, we exploit the fact that, in 1964, the ANES included questions on respondents’
“feeling thermometers” towards different political, demographic, and socioeconomic groups,
with higher values reflecting warmer feelings towards a group. 2SLS estimates in Table A.24
show that Black inflows were positively associated with feelings towards Democrats (column
1) and labor unions (column 2). The 1940-1960 change in the Black share was also positively
correlated with whites’ feelings thermometers towards African Americans (column 3) and the
43Several papers have documented that scapegoating and violence against minorities are more likely to happenduring times of hardship (Grosfeld et al., 2020; Oster, 2004; Voigtlander and Voth, 2012).
44We predict economic growth using a simple Bartik-style approach, interacting 1940 industry shares atthe county level with national growth rates of each industry in each subsequent decade, and averaging acrossindustries (for each decade).
45See Table A.23 for the 2SLS coefficients plotted in Figure A.11.
25
NAACP (column 4), even though these results are not statistically significant.
5.3.3 Residential Segregation and Independent Local Governments
The lack of (between county) white flight and the higher support for civil rights among some
white voters do not imply that white residents welcomed Black migrants into their neighbor-
hoods. As prior research shows, Black migration may have increased racial segregation, and
efforts of whites to avoid sharing public goods with Blacks (Alesina et al., 1999, 2004). We
confirm these patterns in our setting. We document that this mechanism may have amplified,
rather than dampening, the positive effect of the Great Migration on demand for civil rights.
On the one hand, residential segregation was a way to diffuse white animosity caused by Black
migration into white neighborhoods. On the other hand, the ability to segregate may have
sustained and reinforced perceptions among liberal whites that the civil rights legislation was
mostly about the South, and that their privileges would not be significantly eroded.
Table 6, column 1, provides evidence consistent with these ideas by focusing on pro-civil
rights demonstrations. Panel A replicates results in Table 3 (column 6), whereas Panels B
and C split the sample in counties with 1940 residential segregation (constructed using the
procedure in Logan and Parman, 2017) below and above the median, respectively. The F-stat
falls below conventional levels in Panel C, and so results should be interpreted with caution.
However, the pattern that emerges is clear: Black in-migration increased the frequency of pro-
civil rights demonstrations only in counties with higher residential segregation. Said differently,
support for civil rights increased more in counties where inter-group contact in the housing
market was lower.
This may have happened because residents of initially segregated counties had little contact
with Blacks, and were able to further isolate themselves from inter-racial tensions that Black
migration may have brought to neighborhoods and housing markets. To achieve this goal,
whites could create more homogeneous local jurisdictions. In columns 2 to 5 of Table 6, we
examine whether the Great Migration increased the number of local jurisdictions using data
from the Census of Government. We replicate the regressions in column 1 focusing on the
change in the (log of) number of: i) total jurisdictions (column 2); ii) school districts (column
3); iii) special districts (column 4); and, iv) municipalities (column 5). In the full sample
(Panel A), Black in-migration had a positive and statistically significant effect on the number
of local jurisdictions (column 2) – a pattern driven entirely by school districts (column 3).
However, this happened only in counties with residential segregation above the median (Panel
C).
One interpretation, consistent with the historical evidence (Sugrue, 2008), is that, since
higher residential segregation lowered the probability that Black and white pupils shared the
same school district, whites’ incentives to create local jurisdictions were stronger in more segre-
gated counties. Coupled with findings in column 1, this suggests that population sorting within
counties and the creation of independent jurisdictions might have reduced potential backlash by
allowing whites to live in racially homogeneous communities, where the probability of sharing
public goods with Blacks was low.
26
6 Legislators’ Behavior
This section studies the impact of the Great Migration on support for civil rights legislation
among members of the House. First, it shows that Congress members representing CDs that
received more Black migrants were more likely to vote in favor of pro-civil rights bills and to
sign discharge petitions to promote racial equality (Section 6.1). Next, it documents that Black
in-migration increased political polarization on racial issues (Section 6.2).
6.1 Ideology Scores and Discharge Petitions
6.1.1 Ideology Scores
We begin the analysis of legislators’ behavior by focusing on the ideology scores from Bateman
et al. (2017), which take more negative (resp. positive) values for more liberal (resp. con-
servative) voting behavior on civil rights bills. Columns 1 to 3 in Table 7 present results for
the change in agnostic ideology scores, stacking the data for the 78-82 and the 82-88 Congress
periods, reporting 2SLS and first stage coefficients in Panels A and B respectively. Following
Autor et al. (2020) and Bonomi et al. (2020), to deal with the potential issue of mean rever-
sion, in addition to the controls included in our preferred specification above, we also add the
interaction between period dummies and the baseline ideology score of legislators. The point
estimate reported in column 1 (Panel A) is negative, but quantitatively small and imprecisely
estimated.
However, when examining the results separately by Congress period, a more nuanced picture
emerges. Black in-migration had a strong, negative effect on the ideology scores of legislators in
the first Congress period (column 2), and a negligible, positive, and not statistically significant
effect in the second period (column 3). While the F-stat falls below conventional levels in
column 2, suggesting that our estimates should be interpreted with some caution, these findings
indicate that legislators’ ideology moved to the left between Congress 78 and Congress 82, and
did not change significantly afterwards. Results are robust to focusing on the constrained
version of the ideology scores (columns 4 to 6).
In our baseline specification, we map the 1940-1950 (resp. 1950-1960) Black in-migration
to the 78-82 (resp. 82-88) Congress period, so as to both have the longest periods without
redistricting and end the analysis with the Congress that passed the CRA. Appendix C.5 verifies
that our findings are robust to different timing conventions. It also shows that there are no
pre-trends, and runs a number of robustness checks to verify that results are not influenced by
strategic gerrymandering of CD boundaries, possibly induced by Black in-migration (Kaufman
et al., 2017).
6.1.2 Signatures on Discharge Petitions
Due to gatekeeping imposed by southern Democrats, civil rights bills were unlikely to reach the
floor of the House, unless northern legislators were willing to undertake non-standard actions.
Discharge petitions represent the best example of such non-conventional tools at disposal of
non-southern legislators (Pearson and Schickler, 2009). Since there are not enough discharge
27
petitions filed during the 82-88 Congress period, we focus on the 1940s, when several discharge
petitions were filed and signed on the same topics – fair employment legislation (FEPC), the
poll tax, and anti-lynching legislation – both at the beginning and at the end of the decade.
Although all three topics featured prominently in the political debate during the 1940s, some
differences existed between them. First, the salience of both the poll tax and anti-lynching
legislation gradually declined relative to that of anti-discrimination employment legislation
during the 1940s. For instance, the last discharge petition on either the poll tax or anti-
lynching legislation was filed during the 80th Congress, whereas discharge petitions on FEPC
were filed also in the early 1950s (Table A.5). Second, anti-lynching legislation and, to a lesser
extent, the abolition of the poll tax almost exclusively concerned racial relations in the South;
conversely, employment protection legislation had a direct, strong impact not only in the South
but also in the North (Sugrue, 2014). For both reasons, one may expect FEPC to be the most
relevant category, where northern legislators may have tried to signal their (pro-civil rights)
stance the most.
Figure 4 plots the 2SLS point estimate (with 95% confidence intervals) for the effects of
changes in the Black share on the change in the probability of signing a discharge petition on any
of the three topics (first dot) and then on each topic separately (second to fourth dots from the
left). Black in-migration had a positive and statistically significant effect on the probability of
signing a discharge petition on all topics. Consistent with the previous discussion, these effects
are larger and more precisely estimated for FEPC legislation than for other categories.46
Table A.27 presents additional, suggestive evidence estimating a “levels on changes” specifi-
cation. The key limitation of this strategy is that unobservable CD fixed characteristics cannot
be controlled for. However, it allows us to provide evidence also for the second Congress pe-
riod. In columns 1 to 3 (resp. 4 to 6), the dependent variable is the number of signatures
on discharge petitions per legislator signed (resp. the share of petitions signed prior to the
50th signature) during the 78-82 and the 83-88 Congress periods.47 2SLS coefficients show that
Black in-migration was positively associated with a higher number of signatures per legislator.
Moreover, legislators representing CDs that received more African Americans were more likely
to sign petitions earlier in the process – a pattern consistent with higher legislators’ support
for civil rights legislation (Schickler, 2016).
46Table A.25 reports the coefficients associated with Figure 4. Since the KP F-stat is below conventionallevels, results should be interpreted with some caution. The change in the probability of signing a petitionon FEPC, anti-lynching legislation, and the poll tax is taken over Congresses 81 to 78, 80 to 77, and 79 to77 respectively. Since petitions on the three topics were not always signed in the same Congress year andwere not always comparable with each other (see Table A.5), we checked the robustness of our results using anumber of alternative time windows. Reassuringly, they always remained similar to those presented in Figure 4.Results in Figure 4 and in Table A.25 are obtained from a slightly larger number of CDs relative to those forwhich legislators’ ideology is available. Restricting attention to these CDs leaves our results unchanged (seeTable A.26).
47Only one petition was filed during Congress 82, obtaining merely 16 signatures (Table A.5). For this reason,defining the end (resp. the beginning) of the first (resp. second) Congress period with Congress 82 or Congress83 makes no difference.
28
6.2 Heterogeneous Effects
6.2.1 Within vs Between Party Changes
The shift in legislators’ ideology to the left documented in Table 7 might come from two,
non-mutually exclusive forces. First, it may reflect changes taking place between parties,
if Republican legislators were replaced by Democratic legislators. Second, it might capture
changes taking place within parties, if the ideology of Congress members of the same party
shifted towards more liberal positions.
To disentangle these mechanisms, we create four separate dummies for each possible party
transition experienced by a CD between the beginning and the end of a Congress period – from
Republican to Democratic, from Republican to Republican, from Democratic to Democratic,
and from Democratic to Republican. Next, we interact such dummies with the change in the
Black share, and include all interactions in the same specification (in addition to the direct
effect of Black in-migration). The omitted category is the interaction between the change in
the Black share and the Democratic-to-Democratic dummy. Thus, coefficients on the various
interaction terms should be interpreted as the effects of Black in-migration in a CD undergoing
a given transition, as compared to CDs that were Democratic both at the beginning and at
the end of the Congress period.
We report 2SLS estimates from the baseline stacked first difference regression in Table 8, for
the agnostic and the constrained version of the ideology scores in columns 1 and 2 respectively.48
The main effect of Black in-migration is negative and marginally statistically significant (with
a p-value of 0.056). Focusing on the interaction terms, the coefficient on the “Republican
to Democratic” transition is negative and marginally significant, indicating that the Great
Migration produced a more pronounced shift towards the left in CDs that switched from the
Republican to the Democratic Party – evidence of a between party adjustment. Perhaps more
interestingly, the coefficient on the interaction for the “Republican to Republican” transition is
positive and statistically significant. This implies that Black in-migration induced Republican
legislators to become less liberal on racial issues in CDs that were Republican both at the
beginning and at the end of the period. The within party adjustment taking place in GOP
dominated CDs further suggests that the average effects estimated above may mask between
party polarization on racial issues.
6.2.2 Political Polarization
To examine the possibility that the Great Migration increased political polarization, we follow
the approach used in Autor et al. (2020) and Tabellini (2020) for trade and immigration
respectively. We define liberal (resp. moderate) Democrats those legislators with an ideology
score below (resp. above) the median score for Democrats in Congress 78. Likewise, moderate
(resp. conservative) Republicans are defined as Congress members with an ideology score
below (resp. above) the median score for Republicans in Congress 78. Table 8 estimates our
48Since there are now four endogenous regressors and four instruments, we report the AP F-stat associatedwith each instrument, as suggested by Angrist and Pischke (2008). In two out of four cases, these are belowconventional levels, likely due to the highly demanding nature of the exercise.
29
baseline stacked first difference specification, using as dependent variable the change in the
probability of electing a liberal Democrat (column 3), a moderate Democrat (column 4), a
moderate Republican (column 5), and a conservative Republican (column 6).
2SLS coefficients document that Black in-migration had a positive, statistically significant
effect on the probability of electing a liberal Democrat. The remaining coefficients are impre-
cisely estimated, but suggest that this was accompanied by a reduction in the probability of
electing moderate legislators (either Democratic or Republican). If anything, the point esti-
mate for the probability of electing a conservative Republican is positive, albeit small and not
statistically significant.
Section 6.1.1 above showed that the Great Migration had a strong, negative effect on the
ideology scores of legislators in the 1940s, but a positive – and in some specifications statistically
significant (see Table C.13) – one in the 1950s. Motivated by this evidence, in Figure 5, we
re-estimate the regressions reported in columns 3 to 6 of Table 8 separately for each of the
two decades in Panels A and B respectively.49 The pattern arising is rather striking. In the
1940s, Black in-migration had a strong, positive effect on the probability of electing a liberal
Democrat. This was accompanied by a steep reduction in the probability of electing both
moderate Democrats and conservative Republicans. If anything, the probability of electing a
moderate Republican increased with Black inflows, even though results are not statistically
significant. These findings reflect the combination of between and within party adjustments
discussed above.
In the 1950s, instead, Black in-migration was associated with a statistically significant
increase in the probability of electing a conservative Republican, mirrored by the corresponding
decline in the probability of electing a moderate Republican. The effects of Black in-migration
on the probability of electing Democrats with different ideological stances are very small in size
and imprecisely estimated. The right-ward shift of Republican legislators during the 1950s may
have been motivated by strategic considerations, as the GOP tried to win the votes of whites
who were becoming increasingly concerned about the racial mixing of their neighborhoods
(Sugrue, 2014).
Since quantitatively results in Panel A are larger than those in Panel B, on average, leg-
islators’ ideology moved to the left. However, when inspecting these dynamics more carefully,
polarization becomes evident. These findings are also consistent with the possibility that lo-
cal responses to the Great Migration were partly influenced by national considerations: even
though the Democratic Party “lost the South” by openly promoting the civil rights agenda
(Kuziemko and Washington, 2018), this strategy might have been instrumental to win urban
areas of the West and the North. At the same time, the Republican Party might have tried
to strengthen its conservative position at the national level, so as to attract dissatisfied white
voters leaving the Democratic Party in the South.
49Table A.28 reports the 2SLS and first stage coefficients associated with Figure 5.
30
7 Conclusions
The Great Migration of African Americans was one of the largest episodes of internal migration
in American history. Between 1940 and 1970, more than 4 million Blacks left the US South for
northern and western destinations. During this same period, the civil rights movement strug-
gled and eventually succeeded to eliminate formal impediments to Black political participation
and to remove (at least) de jure racial segregation. In this paper, we study the effects that
Black in-migration had both on voters’ demand for racial equality and on legislators’ support
for civil rights legislation.
Using a version of the shift-share instrument (Card, 2001; Boustan, 2010), we show that
Black in-migration increased the Democratic vote share in Congressional elections and raised
the frequency of pro-civil rights demonstrations. Our estimates suggest that these effects were
at least in part due to the behavior of white voters, who also joined civil rights protests.
We corroborate this interpretation using historical survey data and examining patterns of
heterogeneity across counties. Next, we document that Congress members representing CDs
that received more African Americans became more liberal on racial issues, and more actively
supported civil rights legislation. These average effects, however, mask substantial polarization
between parties on racial issues.
Our paper complements the existing literature on the Great Migration, which has, especially
in recent times, emphasized the long run, negative impact that the latter had on both racial
residential segregation and economic mobility for African Americans. Our findings, instead,
paint a more nuanced picture. They indicate that, as predicted by Gunnar Myrdal back in
1944, Black in-migration to the US North and West was instrumental for the development of
the civil rights movement – the social movement that was of historical importance to reduce
political, social, and economic racial inequality in the United States.
When contrasted with other works on the political effects of migration, our results raise
an intriguing set of questions. Under what conditions can migration and inter-group contact
more broadly lead to the formation of cross-group coalitions? When, instead, is backlash from
original residents more likely to prevail? In the specific context of the Great Migration and of
the civil rights movement, our evidence suggests that cross-race cooperation can emerge when
individuals belonging to different groups share similar (in this context, more liberal) values,
or when group members interact in domains, such as the working environment, where the
emergence of a common identity and the existence of common goals can sustain political or
social coalitions. At the same time, contact in other domains, such as residential markets and
a more private sphere, might lead to inter-group conflict and majority backlash.
31
References
Acemoglu, D., U. Akcigit, and M. A. Celik (2020). “Radical and Incremental Innovation: The Roles ofFirms, Managers and Innovators”. American Economic Journal: Macroeconomics. Forthcoming.
Adams, H. S. (1966). “The Dingell-Lesinski 1964 Primary Race”. The Western Political Quarterly 19 (4),688–696.
Adao, R., M. Kolesar, and E. Morales (2019). “Shift-share Designs: Theory and Inference”. The QuarterlyJournal of Economics 134 (4), 1949–2010.
Adler, E. S. (2003). Congressional District Data File, [78]. University of Colorado, Boulder, CO.http://socsci. colorado. edu/˜ esadler/research. htm.
Alesina, A., R. Baqir, and W. Easterly (1999). “Public Goods and Ethnic Divisions”. The Quarterlyjournal of economics 114 (4), 1243–1284.
Alesina, A., R. Baqir, and C. Hoxby (2004). “Political Jurisdictions in Heterogeneous Communities”.Journal of political Economy 112 (2), 348–396.
Allport, G. W. (1954). The Nature of Prejudice. Oxford, England: Addison-Wesley.
Altonji, J. G. and R. M. Blank (1999). “Race and Gender in the Labor Market”. Handbook of laboreconomics 3, 3143–3259.
Aneja, A. and C. F. Avenancio-Leon (2019). “The Effect of Political Power on Labor Market Inequality:Evidence from the 1965 Voting Rights Act”. Working paper.
Angrist, J. D. and J.-S. Pischke (2008). Mostly Harmless Econometrics: An Empiricist’s Companion.Princeton university press.
Arzheimer, K. (2009). “Contextual Factors and the Extreme Right Vote in Western Europe, 1980–2002”.American Journal of Political Science 53 (2), 259–275.
Autor, D., D. Dorn, G. Hanson, and K. Majlesi (2020). “Importing Political Polarization? The ElectoralConsequences of Rising Trade Exposure”. American Economic Review 110 (10), 3139–83.
Autor, David, H. and D. Dorn (2013). “The Growth of Low-Skill Service Jobs and the Polarization of theUS Labor Market”. American Economic Review 103 (5), 1553–97.
Bailer, L. H. (1943). “The Negro Automobile Worker”. Journal of Political Economy 51 (5), 415–428.
Bailer, L. H. (1944). “The Automobile Unions and Negro Labor”. Political Science Quarterly 59 (4),548–577.
Baran, C., E. Chyn, and B. A. Stuart (2020). ”The Great Migration and Educational Opportunity”.Working Paper.
Bateman, D. A., J. D. Clinton, and J. S. Lapinski (2017). “A House Divided? Roll Calls, Polarization,and Policy Differences in the US House, 1877–2011”. American Journal of Political Science 61 (3),698–714.
Bayer, P. and K. K. Charles (2018). “Divergent paths: A New Perspective on Earnings Differencesbetween Black and White Men since 1940”. The Quarterly Journal of Economics 133 (3), 1459–1501.
Bernini, A., G. Facchini, and C. Testa (2018). “Race, Representation and Local Governments in the USSouth: The Effect of the Voting Rights Act”. Discussion paper No. 12774, CEPR.
32
Besley, T., T. Persson, and D. M. Sturm (2010). “Political Competition, Policy and Growth: Theory andEvidence from the US”. The Review of Economic Studies 77 (4), 1329–1352.
Beth, R. S., Government, and F. Division (2003). The Discharge Rule in the House: Principal Featuresand Uses. Congressional Research Service.
Blalock, H. M. (1967). Toward a Theory of Minority-Group Relations, Volume 325. New York: Wiley.
Bonomi, G., N. Gennaioli, and G. Tabellini (2020). “Identity, Beliefs, and Political Conflict”. WorkingPaper.
Borusyak, K., P. Hull, and X. Jaravel (2018). “Quasi-experimental Shift-share Research Designs”. WorkingPaper 24997, NBER.
Bositis, D. A. (2012). Blacks and the 2012 Democratic National Convention. Joint Center for Politicaland Economic Studies.
Boustan, L. P. (2009). “Competition in the Promised Land: Black Migration and Racial Wage Conver-gence in the North, 1940–1970”. The Journal of Economic History 69 (3), 755–782.
Boustan, L. P. (2010). “Was Postwar Suburbanization “White Flight”? Evidence from the Black Migra-tion”. The Quarterly Journal of Economics 125, 417–443.
Boustan, L. P. (2016). Competition in the Promised Land: Black Migrants in Northern Cities and LaborMarkets. Princeton University Press.
Boustan, L. P., D. Bunten, and O. Hearey (2014). “Urbanization in the United States, 1800-2000”.Working paper 19041, NBER.
Bowles, G. K., J. D. Tarver, C. L. Beale, and E. S. Lee (1990). Net Migration of the Population by Age,Sex, and Race, 1950-1970 [computer file]. ICPSR ed., Study (8493).
Cantoni, E. and V. Pons (2020). “Do Interactions with Candidates Increase Voter Support and Partici-pation?: Experimental Evidence from Italy”. Economics & Politics. Forthcoming.
Card, D. (2001). “Immigrant Inflows, Native Outflows, and the Local Labor Market Impacts of HigherImmigration”. Journal of Labor Economics 19 (1), 22–64.
Card, D. and G. Peri (2016). “Immigration Economics by George J. Borjas: a Review Essay”. Journalof Economic Literature 54 (4), 1333–49.
Carmines, E. G. and J. A. Stimson (1989). Issue Evolution: Race and the Transformation of AmericanPolitics. Princeton University Press.
Cascio, E., N. Gordon, E. Lewis, and S. Reber (2010). “Paying for Progress: Conditional Grants and theDesegregation of Southern Schools”. The Quarterly Journal of Economics 125 (1), 445–482.
Cascio, E. U. and E. Washington (2014). “Valuing the Vote: The Redistribution of Voting Rights andState Funds Following the Voting Rights Act of 1965”. The Quarterly Journal of Economics 129 (1),379–433.
Caughey, D., M. C. Dougal, and E. Schickler (2020). “Policy and Performance in the New Deal Realign-ment: Evidence from old data and new methods”. The Journal of Politics 82 (2), 494–508.
Caughey, D. and C. Warshaw (2018). “Policy Preferences and Policy Change: Dynamic Responsivenessin the American States, 1936–2014”. American Political Science Review 112 (2), 249–266.
33
Chetty, R., N. Hendren, M. R. Jones, and S. R. Porter (2020). “Race and Economic Opportunity inthe United States: An Intergenerational Perspective”. The Quarterly Journal of Economics 135 (2),711–783.
Clubb, J. M., W. H. Flanigan, and N. H. Zingale (1990). Partisan Realignment: Voters, Parties, andGovernment in American History. Beverly Hills, CA: SAGE.
Collins, W. J. (2020). “The Great Migration of Black Americans from the US South: A Guide andInterpretation”. Working Paper 27268, NBER.
Collins, W. J. and R. A. Margo (2007). “The Economic Aftermath of the 1960s Riots in American Cities:Evidence from Property Values”. The Journal of Economic History 67 (4), 849–883.
Collins, W. J. and M. H. Wanamaker (2014, January). “Selection and Economic Gains in the GreatMigration of African Americans: New Evidence from Linked Census Data”. American EconomicJournal: Applied Economics 6 (1), 220–52.
Collins, W. J. and M. H. Wanamaker (2015). “The Great Migration in Black and White: New Evidenceon the Selection and Sorting of Southern Migrants”. The Journal of Economic History 75 (4), 947–992.
Dahis, R., E. Nix, and N. Qian (2020). “Choosing Racial Identity in the United States, 1880-1940”.Working Paper 26465, NBER.
Derenoncourt, E. (2018). “Can You Move to Opportunity? Evidence from the Great Migration”. Workingpaper.
Dustmann, C., K. Vasiljeva, and A. P. Damm (2019). “Refugee Migration and Electoral Outcomes”.Review of Economic Studies 86 (5), 2035–2091.
Engstrom, E. (2013). Partisan Gerrymandering and the Construction of American Democracy. Universityof Michigan Press.
Enos, R. D. (2016). “What the Demolition of Public Housing Teaches us about the Impact of RacialThreat on Political Behavior”. American Journal of Political Science 60 (1), 123–142.
Eriksson, K. (2019). “Moving North and into jail? The great migration and black incarceration”. Journalof Economic Behavior & Organization 159, 526–538.
Farber, H. S., D. Herbst, I. Kuziemko, and S. Naidu (2018). “Unions and Inequality over the TwentiethCentury: New Evidence from Survey Data”. Working Paper 24587, NBER.
Feigenbaum, J. J. and A. B. Hall (2015). “How Legislators Respond to Localized Economic Shocks:Evidence from Chinese Import Competition”. The Journal of Politics 77 (4), 1012–1030.
Feigenbaum, J. J., S. Mazumder, and C. B. Smith (2020). “When Coercive Economies Fail: The PoliticalEconomy of the US South After the Boll Weevil”. Working Paper 27161, NBER.
Feinstein, B. D. and E. Schickler (2008). “Platforms and Partners: The Civil Rights Realignment Recon-sidered”. Studies in American Political Development 22 (1), 1.
Fouka, V., S. Mazumder, and M. Tabellini (2018). “From Immigrants to Americans: Race and Assimila-tion during the Great Migration”. Harvard Business School BGIE Unit Working Paper .
Frymer, P. and J. M. Grumbach (2020). “Labor Unions and White Racial Politics”. American Journalof Political Science.
34
Gardner, J. and W. Cohen (1992). “Demographic Characteristics of the Population of the United States,1930-1950: County-Level”. Ann Arbor, MI: Inter-university Consortium for Political and Social Re-search [distributor], 1992-02-16. https://doi.org/10.3886/ICPSR00020.v1 .
Gentzkow, M. (2016). “Polarization in 2016”. Toulouse Network for Information Technology Whitepaper .
Gibson, C. and K. Jung (2005). Historical Census Statistics on Population Totals by Race, 1790 to 1990,and by Hispanic Origin, 1970 to 1990, for Large Cities and Other Urban Places in the United States.Citeseer.
Goldsmith-Pinkham, P., I. Sorkin, and H. Swift (2020). “Bartik Instruments: What, When, Why, andHow”. American Economic Review 110 (8), 2586–2624.
Grant, K. (2020). Relocation & Realignment: How the Great Migration Changed the Face of the Demo-cratic Party. Temple University Press.
Gregory, J. N. (1995). “The Southern Diaspora and the Urban Dispossessed: Demonstrating the CensusPublic Use Microdata Samples”. The Journal of American History 82 (1), 111–134.
Gregory, J. N. (2006). The Southern Diaspora: How the Great Migrations of Black and White SouthernersTransformed America. University of North Carolina Press.
Gregory, J. N. and J. Estrada (2019). “NAACP History and Geography. Mapping American SocialMovement”.
Gregory, J. N. and A. Hermida (2019). “Congress of Racial Equality (CORE) Actions, 1942-1972. MappingAmerican Social Movement”.
Grosfeld, I., S. O. Sakalli, and E. Zhuravskaya (2020). “Middleman Minorities and Ethnic Violence:Anti-Jewish pogroms in the Russian Empire”. The Review of Economic Studies 87 (1), 289–342.
Grossman, J. R. (1991). Land of Hope: Chicago, Black Southerners, and the Great Migration. Universityof Chicago Press.
Grove, W. A. and C. Heinicke (2003). “Better Opportunities or Worse? The Demise of Cotton HarvestLabor, 1949–1964”. The Journal of Economic History 63 (3), 736–767.
Haines, M. R. et al. (2010). “Historical, Demographic, Economic, and Social Data: the United States,1790–2002”. Ann Arbor, MI: Inter-university Consortium for Political and Social Research.
Inglehart, R. (2015). The Silent Revolution: Changing Values and Political Styles among Western Publics.Princeton University Press.
Jaeger, D. A., J. Ruist, and J. Stuhler (2018). Shift-share instruments and the impact of immigration.Working paper 24285, NBER.
Jones, D. B. and R. Walsh (2018). “How Do Voters Matter? Evidence from US Congressional Redistrict-ing”. Journal of Public Economics 158, 25–47.
Katznelson, I. and J. S. Lapinski (2006). “The Substance of Representation: Studying Policy Contentand Legislative Behavior”. In The Macropolitics of Congress, pp. 96–126. Princeton University PressPrinceton, NJ.
Kaufman, A., G. King, and M. Komisarchik (2017). “How to Measure Legislative District Compactnessif you only know it When you see it”. American Journal of Political Science.
35
Keyssar, A. (2009). The Right to Vote: The Contested History of Democracy in the United States. BasicBooks.
Kousser, J. M. (2010). “The Immutability of Categories and The Reshaping of Southern Politics”. AnnualReview of Political Science 13, 365–383.
Kroth, V., V. Larcinese, and J. Wehner (2016). “A Better Life for All? Democratization and Electrificationin post-apartheid South Africa”. The Journal of Politics 78 (3), 774–791.
Kuziemko, I. and E. Washington (2018). “Why Did the Democrats Lose the South? Bringing New Datato an Old Debate”. American Economic Review 108 (10), 2830–67.
Lawson, S. F. (1976). Black Ballots: Voting Rights in the South, 1944-1969. Columbia University Press.
Logan, T. D. and J. M. Parman (2017). “The National Rise in Residential Segregation”. The Journal ofEconomic History 77 (1), 127–170.
Lott, Jr, J. R. and L. W. Kenny (1999). “Did Women’s Suffrage Change the Size and Scope of Govern-ment?”. Journal of political Economy 107 (6), 1163–1198.
Lowe, M. (2017). “Types of Contact: A Field Experiment on Collaborative and Adversarial CasteIntegration”. Behavioral and Experimental Economics eJournal .
Margo, R. A. (1991). “Segregated Schools and the Mobility Hypothesis: a Model of Local GovernmentDiscrimination”. The Quarterly Journal of Economics 106 (1), 61–73.
McAdam, D. (1982). Political Process and the Development of Black Insurgency, 1930-1970. Chicago andLondon: University of Chicago Press.
Mian, A., A. Sufi, and F. Trebbi (2010). “The Political Economy of the US Mortgage Default Crisis”.American Economic Review 100 (5), 1967–98.
Miles, S. (2000). Youth Lifestyles in a Changing World. McGraw-Hill Education (UK).
Miller, G. (2008). “Women’s Suffrage, Political Responsiveness, and Child Survival in American History”.The Quarterly Journal of Economics 123 (3), 1287–1327.
Moon, H. L. (1948). Balance of Power: The Negro Vote. Greenwood Publishing Group.
Muller, C. (2012). “Northward Migration and the Rise of Racial Disparity in American Incarceration,1880–1950”. American Journal of Sociology 118 (2), 281–326.
Myrdal, G. (1944). An American Dilemma: the Negro Problem and Modern Democracy. Harper andRow, Publishers.
Neal, D. A. and W. R. Johnson (1996). “The Role of Premarket Factors in Black-White Wage Differences”.Journal of political Economy 104 (5), 869–895.
Oster, E. (2004, March). “Witchcraft, Weather and Economic Growth in Renaissance Europe”. Journalof Economic Perspectives 18 (1), 215–228.
Parker, K., J. Horowitz, and B. Mahl (2016). On Views of Race and Inequality, Blacks and Whites areWorlds Apart: About Four-in-ten Blacks are Doubtful that the US Will Ever Achieve Racial Equality.Pew Research Center.
Pearson, K. and E. Schickler (2009). “Discharge Petitions, Agenda Control, and the Congressional Com-mittee System, 1929–76”. The Journal of Politics 71 (4), 1238–1256.
36
Peri, G. and C. Sparber (2011). “Assessing Inherent Model Bias: An Application to Native Displacementin Response to Immigration”. Journal of Urban Economics 69 (1), 82–91.
Pons, V. (2018). “Will a Five-minute Discussion Change Your Mind? A Countrywide Experiment onVoter Choice in France”. American Economic Review 108 (6), 1322–63.
Poole, K. T. and H. Rosenthal (1985). “A Spatial Model for Legislative Roll Call Analysis”. AmericanJournal of Political Science 29 (2), 357–384.
Qian, N. and M. Tabellini (2020). “Discrimination, Disenfranchisement and African American WWIIMilitary Enlistment‘”.
Rao, G. (2019). “Familiarity does not breed contempt: Generosity, Discrimination, and Diversity in DelhiSchools”. American Economic Review 109 (3), 774–809.
Reber, S. J. (2011). “From Separate and Unequal to Integrated and Equal? School Desegregation andSchool Finance in Louisiana”. The Review of Economics and Statistics 93 (2), 404–415.
Reny, T. T. and B. J. Newman (2018). “Protecting the Right to Discriminate: The Second Great Migrationand Racial Threat in the American West”. American Political Science Review 112 (4), 1104–1110.
Ruggles, S., K. Genadek, R. Goeken, J. Grover, and M. Sobek (2015). Integrated Public Use MicrodataSeries: Version 6.0 [dataset]. Minneapolis: University of Minnesota.
Schickler, E. (2016). Racial Realignment: The Transformation of American Liberalism, 1932–1965.Princeton University Press.
Sequeira, S., N. Nunn, and N. Qian (2020). “Immigrants and the Making of America”. Review of EconomicStudies 87 (1), 382–419.
Shertzer, A. and R. P. Walsh (2019). “Racial Sorting and the Emergence of Segregation in AmericanCities”. Review of Economics and Statistics 101 (3), 1–14.
Smith, J. P. and F. R. Welch (1989). “Black Economic Progress After Myrdal”. Journal of economicliterature 27 (2), 519–564.
Snyder, J. M. and S. Ansolabehere (2008). The End of Inequality: One Person, One Vote and theReshaping of American Politics. WW Norton and Company.
Steinmayr, A. (2020). “Contact versus Exposure: Refugee Presence and Voting for the Far-Right”. Reviewof Economics and Statistics, 1–47.
Sugrue, T. J. (2008). Sweet Land of Liberty: The Forgotten Struggle for Civil Rights in the North. RandomHouse.
Sugrue, T. J. (2014). The Origins of the Urban Crisis: Race and Inequality in Postwar Detroit-UpdatedEdition. Princeton University Press.
Swift, E. K., R. G. Brookshire, D. T. Canon, E. C. Fink, J. R. Hibbing, B. D. Humes, M. J. Malbin, andK. C. Martis (2000). Database of Congressional Historical Statistics [computer file]. Ann Arbor, MI:Inter-university Consortium for Political and Social Research.
Tabellini, M. (2018). “Racial Heterogeneity and Local Government Finances: Evidence from the GreatMigration”. Working Paper No. 19-006, Harvard Business School BGIE Unit.
Tabellini, M. (2020). “Gifts of The Immigrants, Woes of The Natives: Lessons from The Age of MassMigration”. Review of Economic Studies 87 (1), 454–486.
37
Tiebout, C. M. (1956). “A Pure Theory of Local Expenditures”. Journal of political economy 64 (5),416–424.
Tolbert, C. M. and M. Sizer (1996). US Commuting Zones and Labor Market Areas: A 1990 Update.Technical report.
Trende, S. (2012). The Lost Majority: Why the Future of Government Is Up for Grabs-and Who WillTake It. St. Martin’s Press.
Troy, L. (1957). “Distribution of Union Membership among the States, 1939 and 1953”. NBER Books.
Voigtlander, N. and H.-J. Voth (2012). “Persecution Perpetuated: the Medieval Origins of anti-SemiticViolence in Nazi Germany”. The Quarterly Journal of Economics 127 (3), 1339–1392.
Wasow, O. (2020). “Agenda Seeding: How 1960s Black Protests Moved Elites, Public Opinion andVoting”. American Political Science Review , 1–22.
Whatley, W. C. (1985). “A History of Mechanization in the Cotton South: the Institutional Hypothesis”.The Quarterly Journal of Economics 100 (4), 1191–1215.
Wheaton, B. (2020). “Laws, Beliefs, and Backlash”. Unpublished Working Paper.
Wilkerson, I. (2011). The Warmth of Other Suns: the Epic Story of America’s Great Migration. Vintage.
Wright, G. (2013). Sharing the Prize. Harvard University Press.
Zieger, R. H. (2000). The CIO, 1935-1955. University of North Carolina Press.
38
Figures and Tables
Figure 1. Change in the Black Share across US Counties, 1940 to 1970
Notes: The map plots the change in the share of Blacks in the population between 1940and 1970 for the non-southern counties in our sample.
Figure 2. Heterogeneity by County Characteristics - Social and cultural forces
Notes: The bars report the marginal effect of changes in the Black share (withcorresponding 95% confidence intervals) on the change in the Democratic vote sharefor counties with each 1940 variable above (resp. below) the sample median inorange (resp. blue). Section 5.3.2 describes how each variable is constructed. SeeTable A.21 for the coefficients and standard errors corresponding to the graph.
39
Figure 3. Heterogeneity by County Characteristics - Political and economic forces
Notes: The bars report the marginal effect of changes in the Black share (withcorresponding 95% confidence intervals) on the change in the Democratic vote sharefor counties with each 1940 variable above (resp. below) the sample median inorange (resp. blue). Section 5.3.2 describes how each variable is constructed. SeeTable A.22 for the coefficients and standard errors corresponding to the graph.
Figure 4. Change in Signatures on Discharge Petitions
Notes: The figure plots the 2SLS coefficient (with corresponding 95% confidenceintervals) for the effects of the 1940-1950 change in the Black share on the corre-sponding change in the number of signatures on discharge petitions per legislator.The first dot on the left (“All”) includes discharge petitions on employment protec-tion legislation (FEPC), to promote anti-lynching legislation, and to abolish the polltax. The three remaining dots refer to each of the three issues. Results (both forOLS and 2SLS) and details on the specification are reported in Table A.25.
40
Figure 5. Black in-Migration and Political Polarization
Panel A: 1940s
Panel B: 1950s
Notes: Each bar reports 2SLS coefficients (with corresponding 95% confi-dence intervals) for the effect of changes in the Black share on the changein the probability of electing a member of the House with the correspondingpolitical orientation between Congress 78 and Congress 82 (Panel A) andbetween Congress 82 and Congress 88 (Panel B). The ideology indicatorsare defined in the main text (Section 6.2.2).
41
Table 1. Summary Statistics
Variables Mean Median St. Dev. Min Max Obs
Panel A. 1940 Levels
Black Share (County) 3.60 2.10 0.042 0 46.50 1,139
Black Share (CD) 6.80 7.20 0.047 0 25.40 286
Democratic Vote Share 45.757 49.10 14.564 0 85.00 1,139
Turnout 68.938 69.20 8.479 23.00 97.60 1,139
Civil Rights Scores -0.872 -0.811 0.712 -2.008 1.431 286
Panel B. Changes
Black Share (County) 1.789 0.87 2.303 -4.93 7.537 3,418
Black Share (CD) 5.243 6.686 2.450 -0.808 10.205 571
Democratic Vote Share 1.695 1.80 6.189 -19.267 52.17 3,418
Turnout -6.409 -6.433 3.33 -19.20 9.70 3,418
Civil Rights Scores 0.068 0.065 0.375 -1.177 1.284 571
Notes: The sample includes a panel of the 1,139 non-southern US counties (see Table A.1 for our definition ofsouthern states) for which electoral returns in Congressional elections are available for all Census years between1940 and 1970, and with at least one African American resident in 1940. When relevant, county variablesare collapsed at the Congressional District level, fixing boundaries to Congress 78 as explained in the text.Democratic vote share and turnout refer to Congressional elections, and civil rights scores are the ideologyscores from Bateman et al. (2017). Panel A presents 1940 values, while Panel B reports decadal changes foreach of the variables.
42
Table 2. Congressional Elections
(1) (2) (3) (4) (5) (6) (7)
OLS OLS OLS 2SLS 2SLS 2SLS 2SLS
Panel A. Democratic Vote Share (1940 Mean: 45.89)
Change Black 0.628∗∗∗ 0.607∗∗∗ 0.769∗∗∗ 0.839∗∗∗ 1.165∗∗∗ 1.650∗∗∗ 2.062∗∗∗
Share (0.099) (0.107) (0.131) (0.144) (0.197) (0.286) (0.476)
Panel B. Turnout (1940 Mean: 68.95)
Change Black -0.262∗∗∗ -0.292∗∗∗ -0.268∗∗∗ 0.020 0.194 0.390∗ 0.883∗∗
Share (0.098) (0.092) (0.086) (0.140) (0.172) (0.235) (0.371)
Panel C. First Stage
Predicted Change Black 1.186∗∗∗ 1.357∗∗∗ 1.148∗∗∗ 0.778∗∗∗
Share (0.267) (0.308) (0.311) (0.245)
Specification FD FD FD FD FD FD LD
1940 Black Share X X X X X
1940 Dem incumbent X X X
F-stat 19.69 19.36 13.65 10.13
Observations 3,418 3,418 3,418 3,418 3,418 3,418 1,138
Notes: The sample includes the 1,139 non-southern US counties (see Table A.1 for the definition of southernstates) for which electoral returns in Congressional elections are available for all Census years between 1940and 1970, and with at least one African American resident in 1940. The table reports stacked first differenceregressions in columns 1 to 6, and long difference regressions in column 7. The dependent variable is the decadalchange in the Democratic vote share (resp. turnout) in Congressional elections in Panel A (resp. Panel B).Panel C reports the first stage associated with 2SLS regressions. Columns 1 to 3 estimate equation (1) in thetext with OLS, while remaining columns report 2SLS estimates. The main regressor of interest is the change inthe Black share, which is instrumented with the shift-share instrument described in equation (2) in the text fromcolumn 4 onwards. All regressions are weighed by 1940 county population, and control for state by period fixedeffects. 1940 Black share (resp 1940 Dem dummy) refers to interactions between period dummies and the 1940Black share (resp. a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicansvote share). F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the countylevel, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
43
Table 3. CORE Demonstrations
Dep. Variable Change in 1[Pro-Civil Rights Demonstration]
(1) (2) (3) (4) (5) (6) (7)
OLS OLS OLS 2SLS 2SLS 2SLS 2SLS
Panel A. Main Estimates
Change Black 0.036∗∗∗ 0.027∗∗∗ 0.025∗∗∗ 0.068∗∗∗ 0.048∗∗∗ 0.047∗∗∗ 0.022∗
Share (0.006) (0.006) (0.006) (0.010) (0.009) (0.011) (0.012)
Panel B. First Stage
Predicted Change 1.186∗∗∗ 1.357∗∗∗ 1.148∗∗∗ 1.148∗∗∗
Black Share (0.267) (0.308) (0.311) (0.311)
1940 Black share X X X X X
1940 Dem incumbent X X X
White participants X
F-stat 19.69 19.36 13.65 13.65
Observations 3,418 3,418 3,418 3,418 3,418 3,418 3,418
Notes: The sample includes the 1,139 non-southern US counties (see Table A.1 for the definition of southernstates) for which electoral returns in Congressional elections are available for all Census years between 1940 and1970, and with at least one African American resident in 1940. The dependent variable is the change in theprobability of non-violent demonstrations in support of civil rights coordinated by the CORE. Columns 1 to 3estimate equation (1) in the text with OLS, while remaining columns report 2SLS estimates. The main regressorof interest is the change in the Black share, which is instrumented with the shift-share instrument described inequation (2) in the text from column 4 onwards. All regressions are weighed by 1940 county population, andcontrol for state by period fixed effects. 1940 Black share (resp 1940 Dem dummy) refers to interactions betweenperiod dummies and the 1940 Black share (resp. a dummy equal to 1 if the Democratic vote share in 1940 washigher than the Republicans vote share). Column 7 includes only those demonstrations that were joined by atleast some white participants. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clusteredat the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
44
Table 4. Black in-Migration and Changes in White Population
Dep. Variable Change in White population
(1) (2) (3) (4)
Panel A. 2SLS
Change Black 0.343 0.318 0.278 0.191
population (0.435) (0.428) (0.410) (0.445)
Panel B. First stage
Predicted Change Black 2.430∗∗∗ 2.442∗∗∗ 2.456∗∗∗ 2.396∗∗∗
population (0.352) (0.339) (0.324) (0.340)
F-stat 47.75 52.01 57.30 49.82
Observations 3,418 3,418 1,712 3,418
Baseline Controls X X X
High Urban X
Urban Share X
Avg. Change Black pop. 2,314 2,314 4,456 2,314
Avg. 1940 White pop. 71,001 71,001 120,557 71,001
Avg. Change White pop. 12,058 12,058 19,860 12,058
Notes: The sample is a panel of the 1,139 non-southern US counties (see Table A.1 for our definition of southernstates) for which electoral returns in Congressional elections are available for all Census years between 1940 and1970, and with at least one African American resident in 1940. The table estimates stacked first differenceregressions, reporting 2SLS and first stage results in Panels A and B respectively. The dependent variable is thedecadal change in the white population in the county. The main regressor of interest is the change in the Blackpopulation in the county, instrumented with the shift-share instrument described in equation (2) in the text. Allregressions control for state by period fixed effects. Columns 2 to 4 further include interactions between perioddummies and: i) the 1940 Black share; and ii) a dummy equal to 1 if the Democratic vote share in 1940 washigher than the Republicans vote share). Column 3 restricts attention to counties with 1940 urban share of thepopulation above the sample median (0.356). Column 4 replicates column 2 by including interactions betweenperiod dummies and the 1940 urban share of the population. F-stat is the KP F-stat for weak instruments.Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05,∗ p< 0.1.
45
Table 5. Whites’ Most Important Problem (ANES)
Dep. Variable 1[Pro Civil Rights: Most Important Problem]
(1) (2) (3) (4) (5) (6)
OLS OLS OLS OLS 2SLS 2SLS
Panel A. Main estimates1[Pro Segregation] -0.078∗∗∗
(0.023)
1[Against School -0.078∗∗∗
Integration] (0.019)
1[Against Housing/Work -0.077∗∗
Integration] (0.031)
Change Black Share 0.010 0.017∗∗ 0.022∗∗
(0.010) (0.008) (0.011)
Panel B. First StagePredicted Change 2.184∗∗∗ 2.270∗∗∗
Black Share (0.321) (0.340)
Geography FE State State State Region Region Region
State Controls X X X
Non-Movers X
F-stat 46.17 44.57
Observations 1,570 1,407 1,423 1,602 1,602 927
Mean Dep. Var. 0.105 0.105 0.105 0.105 0.105 0.110
Notes: The sample is restricted to white ANES respondents living in the US North in years 1960 and 1964. Thedependent variable is a dummy equal to 1 if the respondent reports that supporting civil rights is among themost important issues facing the country at the time of the interview (see online appendix D for exact wordingand additional details on the construction of the variable). The regressor of interest in column 2 (resp. column3) is a dummy equal to 1 if the respondent is against the integration of schools (resp. of working environmentand housing). Pro-segregation in column 1 is a dummy if the respondent is either against school integration oragainst working-housing integration. Change Black share (columns 4 to 6) is the change in the Black share atthe state level between 1940 and 1960. Column 4 reports OLS estimates, while columns 5 and 6 present 2SLSestimates, instrumenting the change in the Black share with the predicted number of Black migrants over 1940state population. All regressions include survey year fixed effects and individual controls of respondents (gender,age and education fixed effects and marital status). Columns 1 to 3 control for state fixed effects, while columns4 to 6 control for region fixed effects and 1940 state characteristics (Black share; Democratic incumbency inCongressional elections; share in manufacturing; share of workers in the CIO; urban share). Column 6 restrictsthe sample to respondents who were born in the same state where they lived at the time of the interview. Thebottom row reports the average of the dependent variable. F-stat in columns 5 and 6 is the K-P F-stat for weakinstrument. Panel B reports the first stage. Robust standard errors, clustered at the state level, in parentheses.∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
46
Table 6. Black in-Migration, Residential Segregation, and Local Governments
Dep. Variable Change in Change Log(Number of local governments)
1[Pro-civil Total School Special Municipalities
rights protest] districts districts
(1) (2) (3) (4) (5)
Panel A. Full sampleChange Black Share 0.047*** 0.052*** 0.103*** 0.000 0.008
(0.011) (0.010) (0.017) (0.023) (0.009)
KP F-stat 13.65 13.41 13.41 13.41 13.41
Observations 3,418 3,399 3,399 3,399 3,399
Panel B. Residential Segregation Below MedianChange Black Share -0.004 0.014 -0.050 -0.056 0.037
(0.017) (0.051) (0.086) (0.110) (0.026)
KP F-stat 14.66 14.66 14.66 14.66 14.66
Observations 1,445 1,445 1,445 1,445 1,445
Panel C. Residential Segregation Above MedianChange Black Share 0.053** 0.048*** 0.109*** -0.028 0.018
(0.023) (0.015) (0.025) (0.026) (0.012)
KP F-stat 7.524 7.555 7.555 7.555 7.555
Observations 1,443 1,428 1,428 1,428 1,428
Notes: Panel A estimates the baseline specification with 2SLS (with interactions between year dummies and: i) 1940 Black share;ii) 1940 Democratic incumbency dummy; iii) state fixed effects). Panel B (resp. C) estimates the same set of regressions for countieswith residential segregation below (resp. above) the sample median. Residential segregation refers to the index constructed inLogan and Parman (2017). The table reports 2SLS results replicating the baseline specification. All regressions are weighed by 1940population. Standard errors, clustered at the county level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
47
Table 7. Changes in Legislators’ Ideology
Dep. Variable Change in Civil Rights Ideology (Lower values = more liberal ideology)
Agnostic Scores Constrained Scores
(Baseline mean: -0.872) (Baseline mean: -0.853)
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black Share -0.055 -0.307** 0.047 -0.058 -0.345*** 0.059
(0.042) (0.121) (0.059) (0.043) (0.130) (0.063)
Panel B. First Stage
Predicted Change 1.472*** 1.006*** 1.809*** 1.455*** 1.002*** 1.783***
Black Share (0.441) (0.379) (0.555) (0.445) (0.378) (0.562)
F-stat 11.14 7.068 10.60 10.71 7.038 10.08
Observations 571 286 285 571 286 285
Congress Period 78-82; 82-88 78-82 82-88 78-82; 82-88 78-82 82-88
Notes: The dependent variable is the change in the civil rights ideology scores from Bateman et al. (2017) –“Agnostic” scores in columns 1 to 3, and “Constrained” scores in columns 4 to 6. Lower (resp. higher) valuesof the score refer to more liberal (resp. conservative) ideology (see also Bateman et al., 2017, for more details).Columns 1 and 4 (resp. 2-3, and 5-6) estimate stacked first difference regressions (resp. first difference regressionsfor Congress period 78-82 and 82-88). Panel A reports 2SLS results, while Panels B presents first stage estimates.All regressions are weighed by 1940 congressional district population and control for state by year fixed effectsand include interactions between year dummies and: i) the 1940 Black share in the congressional district; ii) adummy for Democratic incumbency in the 78th Congress in the district; and iii) the ideology score in the districtin the 78th Congress. First difference regressions do not include interactions with year dummies since these areautomatically dropped. F-stat refers to the K-P F-stat for weak instrument. Robust standard errors, clusteredat the congressional district level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
48
Table 8. Heterogeneous Effects on Legislators’ Ideology
Dep. Variable Change in civil rights ideology Change in ideology indicator of elected Congress members
Agnostic Constrained Liberal Moderate Moderate Conservative
Score Score Democrat Democrat Republican Republican
(1) (2) (3) (4) (5) (6)
Change Black Share -0.095* -0.096* 0.081** -0.078 -0.025 0.026
(0.050) (0.050) (0.039) (0.053) (0.031) (0.035)
Change Black -0.011 -0.014
Share*DR (0.033) (0.033)
Change Black 0.058** 0.056**
Share*RR (0.025) (0.025)
Change Black -0.072* -0.071*
Share*RD (0.038) (0.036)
F-stat (Main) 5.98 6.02 11.14 11.14 11.14 11.14
F-stat (DR) 7.13 7.02
F-stat (RR) 13.35 13.67
F-stat (RD) 14.68 16.14
Observations 567 567 571 571 571 571
Notes: The dependent variable is the change in the civil rights ideology score from Bateman et al. (2017) in columns 1 and 2,and the change in the ideology indicator of the Congress member in office in columns 3 to 6. The ideology indicators are definedas: i) liberal (resp. moderate) Democrat if the legislator’s score was below (resp. above) the median score of the Democratic Partymembers in the 78th Congress; ii) moderate (resp. conservative) Republican if the legislator’s score was below (resp. above) themedian score of the Republican Party members in the 78th Congress. Columns 1 and 2 present 2SLS results where the changein the Black share is interacted with dummies for a party transition during the Congress period. The omitted category is that ofdistricts remaining Democratic throughout the period. F-stats in columns 1 and 2 refer to the Angrist and Pischke F-stat for weakinstruments in each first stage, and to the K-P F-stat for weak instruments in columns 3 to 6. All regressions are weighed by 1940congressional district population and control for state by year fixed effects and include interactions between year dummies and: i)the 1940 Black share in the congressional district; ii) a dummy for Democratic incumbency in the 78th Congress in the district; andiii) the ideology score in the district in the 78th Congress. Robust standard errors, clustered at the congressional district level, inparentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
49
A Appendix – Additional Figures and Tables
Figure A.1. Black Emigration Rates from the South, by Decade
Notes: The figure plots the Black emigration rate from the US South for each decade.Source: Adapted from Boustan (2016).
Figure A.2. Northern Legislators Supporting Civil Rights Bills, by Party
Notes: Blue (resp. red) bars plot the share of Democrat (resp. Republican) membersof Congress in the non-South US voting in favor of bills in support of civil rightsbetween the 78th and the 88th Congresses. The first two bars refer to the averagebetween the 78-82 and the 83-88 periods, while the remaining bars display resultsfor each Congress period separately. The 9 bills on the civil rights voted upon inCongress between the 78th and the 88th Congress are listed in Table A.2.
50
Figure A.3. Overall Support for Civil Rights Bills, by Party
Notes: Blue (resp. red) bars plot the share of Democrat (resp. Republican) membersof US Congress voting in favor of bills in support of civil rights between the 78th andthe 88th Congresses. The first two bars refer to the average between the 78-82 andthe 83-88 periods, while the remaining bars display results for each Congress periodseparately. The 9 bills on the civil rights voted upon between the 78th and the 88th
Congress are listed in Table A.2.
Figure A.4. Discharge Petitions on Civil Rights Signed by Northern Legislators
Notes: Blue (resp. red) bars plot the share of Democrat (resp. Republican) membersof Congress in the non-South US signing discharge petitions in favor of civil rightsbills between the 78th and the 88th Congresses. The first two bars refer to theaverage between the 78-82 and the 83-88 periods, while the remaining bars displayresults for each of the two Congress periods separately.
51
Figure A.5. Black Share in 1940
Notes: The map plots the 1940 share of Blacks (divided by county population) forthe non-southern counties in our sample.
Figure A.6. Black Share in Northern Counties, 1940 vs 1970
Notes: Black share of the population for selected non-southern counties in 1940(light blue) and in 1970 (Black). Source: Authors’ calculation from IPUMS data.
52
Figure A.7. Share of Southern Born Blacks in Northern Counties, 1940
Notes: Share of African Americans born in selected southern states living in non-southern counties in 1940. Source: Authors’ calculation from IPUMS data.
Figure A.8. Frequency of CORE Demonstrations, by Type
Notes: The figure plots the number of CORE demonstrations as share of all events oc-curring in our sample period, for each of the four main categories described in the text.The portion of the bar filled with oblique lines (resp. dots) refers to the share of eventsof each category that involved local (resp. national) issues.
53
Figure A.9. Effects of Black in-Migration on CORE Demonstrations, by Type
Notes: The figure plots the 2SLS coefficient (with corresponding 95% confidence intervals)for the effects of the change in the Black share on the change in CORE demonstrations.The first dot from the left considers all demonstrations in our sample; the next four dotsrefer to each specific cause, reported on the x-axis;the last two dots on the right refer todemonstrations that involved, respectively, local and national issues. All regressions areweighed by 1940 population, and include interactions between year dummies and: i) statefixed effects; ii) the 1940 Black share; and iii) a dummy equal to 1 if the Democratic voteshare in 1940 was higher than the Republican vote share. The corresponding estimatesare reported in Table A.7.
54
Figure A.10. Scenarios on Black Behavior and Implied Number of White Switchers
Notes: The red markers indicate the number of votes for the DemocraticParty implied by the 2SLS point estimate in column 6 of Table 2 in a countywith average native-born population and average change in the share ofBlacks. Bars indicate votes cast for the Democratic Party by different seg-ments of the population under different assumptions about turnout and vot-ing preferences of Blacks. In each scenario, “whites” indicates the numberof votes cast for the Democratic Party if white voters did not change theirbehavior in response to Black inflows. Scenario 1 : turnout is 70% for north-ern voters, all Black migrants voted, 70% of northern Blacks and all Blackmigrants voted Democrats. Scenario 2 : Black turnout rate is 70%, 70% ofnorthern Black voters and all Black migrants voted for Democrats. Scenario3 : Black turnout rate is 70%, 70% of Blacks (both migrants and northernresidents) voted for Democrats. Scenario 4 : Black turnout rate is 50%, 70%of Blacks (both migrants and northern residents) voted for Democrats.
55
Figure A.11. Probability that Civil Rights is Most Important Issue for Whites
Notes: The figure plots 2SLS coefficients (with corresponding 95% confidence intervals) for aregression where the dependent variable is the “pro civil rights” dummy reported in Table 5.The first two bars refer to respondents who are not and who are union members (light anddark colors respectively); the third and fourth (resp. fifth and sixth) bars restrict attentionto individuals who are not and who are Democrats (resp. Republicans), with darker colorsreferring to members of the group.
56
Table A.1. List of Southern States
Alabama North Carolina
Arkansas Oklahoma
Florida South Carolina
Georgia Tennessee
Kentucky Texas
Louisiana Virginia
Mississippi West Virginia
Notes: The table presents the listof southern states considered inour analysis. We follow the Cen-sus definition except for Delawareand Maryland: as Boustan (2010)we assign these to the North, asthey were net recipient of Blackmigrants during this period.
Table A.2. Civil Rights Bills Voted in the House, 1943-1964
Congress Year Bill Number Northern Democrats Northern Republicans
Voting Yes Voting Yes
78 1943 HR-7 0.830 0.795
79 1945 HR-7 0.842 0.697
80 1947 HR-29 0.913 0.982
81 1949 HR-3199 0.942 0.696
81 1950 HR-4453 0.790 0.720
84 1956 HR-627 0.914 0.875
85 1957 HR-6127 0.927 0.843
86 1960 HR-8601 0.843 0.813
88 1964 HR-7152 0.918 0.817
Notes: The table lists the bills voted upon in the House of Representatives between Congress 78 and Congress88. The last two columns report the share of northern Democrats (resp. Republicans) who voted in favor ofeach bill relative to all northern Democrats (resp. Republicans).
57
Table A.3. Discharge Petitions, by Party
Poll Tax Lynching FECP Housing Civil Rights Act Total
Panel A. Congress Period: 78th − 82nd
Share Democrats 0.564 0.552 0.500 0.138 - 0.422
Share Republicans 0.304 0.239 0.132 0.024 - 0.147
Panel B. Congress Period: 83rd − 88th
Share Democrats - - 0.632 - 0.677 0.651
Share Republicans - - 0.043 - 0.175 0.154
Notes: The table presents the share of Democrats and Republicans signing discharge petitions on each topic reported inthe top row for the 78-82 (resp. 83-88) Congresses in Panel A (resp. Panel B). When no discharge petition of a given typewas filed in a congress period, the corresponding entry is left missing. Table A.4 reports additional summary statistics forsignatures on discharge petitions. See Table A.5 for the complete list of discharge petitions (by date and by topic). Source:authors’ calculation from Pearson and Schickler (2009).
Table A.4. Discharge Petitions: Summary Statistics
Variables Mean Median St. Dev. Min Max Obs
Panel A: Discharge Petitions by Issue - Congress Period
Poll Tax Lynching FECP Housing Civil Rights Act Total
78th to 82nd 4 3 5 2 0 14
83rd to 88th 0 0 2 1 5 8
Panel B: Discharge Petitions by Legislator – Summary Statistics
Mean Median St. Dev. Min Max Obs.
78th to 82nd 0.772 0.600 0.553 0 2.333 298
83rd to 88th 0.441 0.385 0.298 0 1.286 298
Notes: Panel A presents the number of discharge petitions filed in the two Congress periods (78-82 and 83-88) by type. Panel B reports the summary statistics for the number of petitions signed per legislator for theCongressional Districts in our sample, in either Congress period.
58
Table A.5. Discharge Petitions by Type and Date
Congress Number Topic Total Signatures
73 14 House Restaurant Desegregation 145
74 32 Lynching 218
75 1 Lynching 75
75 5 Lynching 218
76 10 Lynching 218
76 12 Lynching 59
76 34 Poll Tax 49
77 1 Poll Tax 218
77 3 Lynching 59
77 4 Poll Tax 31
77 15 Lynching 29
78 1 Poll Tax 10
78 3 Poll Tax 219
78 5 Lynching 82
78 18 FEPC 41
79 1 Poll Tax 218
79 3 Lynching 150
79 4 FEPC 187
79 24 Public Accommodation 6
80 2 Poll Tax 41
80 9 Lynching 80
81 7 Housing Discrimination 24
81 20 FEPC 110
81 21 FEPC 100
82 6 FEPC 16
83 4 Public Accommodation 71
83 5 FEPC 72
84 5 Civil Rights Act 148
85 1 Civil Rights Act 105
85 6 Civil Rights Act 3
86 3 Civil Rights Act 214
88 2 Anti-Discrimination 4
88 5 Civil Rights Act 174
91 11 Fair Employment 136
Notes: The table reports the list of all pro-civil rights discharge petitions filed between Congresses 73 and 91.Source: adapted from Pearson and Schickler (2009).
59
Table A.6. Congressional Elections by Decade and Presidential Elections
Dep. Variable Congressional Elections Presidential Elections
(1) (2) (3) (4)
Panel A.Democrat Vote Share
Change Black Share 3.051** 0.461 2.374*** 0.577***
(1.513) (0.317) (0.608) (0.144)
Panel B. Turnout
Change Black Share 1.035 0.033 0.504 0.550**
(0.695) (0.337) (0.490) (0.215)
Panel C. First Stage
Predicted Change 0.787*** 1.175*** 1.412*** 1.148***
Black share (0.244) (0.305) (0.441) (0.311)
F-stat 10.41 14.84 10.27 13.21
Observations 1,139 1,139 1,140 3,416
Decade 1940s 1950s 1960s All
Avg. Change Black Share 1.384 1.994 2.331 1.907
Notes: The sample includes the 1,139 non-southern US counties (see Table A.1 for the definition of southernstates) for which electoral returns in Congressional elections are available for all Census years between 1940 and1970, and with at least one African American resident in 1940. The table replicates column 6 of Table 2 forCongressional elections separately for each decade in columns 1 to 3, and for Presidential elections in column4. The main regressor of interest is the change in the Black share, which is instrumented with the shift-shareinstrument described in equation (2) in the text. All regressions are weighed by 1940 county population, andinclude: i) state fixed effects; ii) the 1940 Black share; and iii) a dummy equal to 1 if the Democratic voteshare in 1940 was higher than the Republicans vote share. In column 4, these controls are interacted with yeardummies. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the county level,in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
60
Table A.7. CORE Demonstrations, by Type
Dep. Variable Cause Relevance
All School/ Housing Police Access to Goods Local NationalCollege Brutality
(1) (2) (3) (4) (5) (6) (7)
Panel A. 2SLS
Change Black Share 0.047*** 0.025* 0.019 0.011 0.049*** 0.053*** 0.020*(0.011) (0.013) (0.012) (0.010) (0.012) (0.011) (0.012)
Panel B. First Stage
Predicted Change Black 1.148*** 1.148*** 1.148*** 1.148*** 1.148*** 1.148*** 1.148***Share (0.311) (0.311) (0.311) (0.311) (0.311) (0.311) (0.311)
F-stat 13.65 13.65 13.65 13.65 13.65 13.65 13.65Observations 3,418 3,418 3,418 3,418 3,418 3,418 3,418
Notes: The sample includes the 1,139 non-southern US counties (see Table A.1 for the definition of southern states)for which electoral returns in Congressional elections are available for all Census years between 1940 and 1970, andwith at least one African American resident in 1940. Panel A (resp. B) reports 2SLS (resp. first stage) estimates.The dependent variable is the decadal change in the probability of a protest in each category occurring over adecade. The main regressor of interest is the change in the Black share, which is instrumented with the shift-shareinstrument described in equation (2) in the text. All regressions are weighed by 1940 county population, and includeinteractions between year dummies and: i) state fixed effects; ii) the 1940 Black share; and iii) a dummy equal to 1if the Democratic vote share in 1940 was higher than the Republican vote share. F-stat is the K-P F-stat for weakinstruments. Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01,∗∗ p< 0.05, ∗ p< 0.1.
61
Table A.8. NAACP Chapters
Dep. Variable 1940-1960 Change in 1[NAACP Chapter]
(1) (2) (3) (4)
OLS OLS 2SLS 2SLS
Panel A. Main Estimates
Change Black Share -0.023∗∗∗ 0.048∗∗∗ -0.028 0.075∗∗
(0.008) (0.17) (0.024) (0.035)
Panel B. First Stage
Predicted Change Black 0.761∗∗∗ 0.609∗∗
Share (0.228) (0.240)
NAACP in 1940 YES NO YES NO
F-stat 11.18 6.438
Observations 1,139 932 1,139 932
Notes: The sample includes the 1,139 non-southern US counties (see Table A.1 for the definition of southernstates) for which electoral returns in Congressional elections are available for all Census years between 1940and 1970, and with at least one African American resident in 1940. The dependent variable is the change(between 1940 and 1960) in the presence of NAACP chapters. Columns 2 and 4 restrict attention to countieswith no NAACP chapter in 1940. Columns 1 and 2 estimate OLS regressions, whereas columns 3 and 4 present2SLS results. The main regressor of interest is the 1940-1960 Change Black Share, and is instrumented withthe shift-share instrument constructed in the text in columns 3 and 4. All regressions are weighed by 1940county population, and include interactions between year dummies and: i) state fixed effects; ii) the 1940 Blackshare; and iii) a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republican voteshare. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the county level, inparenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
62
Table A.9. Congressional Elections (CZ)
(1) (2) (3) (4) (5) (6) (7)OLS OLS OLS 2SLS 2SLS 2SLS 2SLS
Panel A. Democrat Vote Share
Change Black Share 0.672** 0.648 0.821** 0.945** 1.886** 2.759*** 3.076***(0.287) (0.416) (0.393) (0.470) (0.952) (1.052) (1.101)
Panel B. Turnout
Change Black Share -0.453*** -0.346* -0.287 -0.452*** 0.361 0.673 0.754(0.132) (0.177) (0.183) (0.171) (0.442) (0.569) (0.633)
Panel C. First Stage
Predicted Change Black 1.033*** 0.755*** 0.671*** 0.679***Share (0.094) (0.176) (0.171) (0.181)
Specification FD FD FD FD FD FD LD1940 Black Share X X X X X1940 Dem Incumbent X X X
F-stat 119.5 18.39 15.43 14.16Observations 1,125 1,125 1,125 1,125 1,125 1,125 375
Notes: The table replicates Table 2 by aggregating the unit of analysis to the commuting zone (CZ). Thedependent variable is the change in the Democratic vote share in Congressional elections (resp. turnout) in PanelA (resp. in Panel B). Panel C reports first stage coefficients. Columns 1 to 3 estimate OLS regressions, whileremaining columns report 2SLS estimates. The main regressor of interest is the change in the Black share, whichis instrumented with the shift-share instrument described in equation (2) in the text from column 4 onwards.All regressions are weighed by 1940 CZ population, and control for state by period fixed effects. 1940 Blackshare (resp. 1940 Dem dummy) refers to interactions between period dummies and the 1940 Black share (resp. adummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicans vote share). F-stat is theK-P F-stat for weak instruments. Robust standard errors, clustered at the CZ level, in parenthesis. Significancelevels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
63
Table A.10. CORE Demonstrations (CZ)
Dep. Variable Change in Pr.(Pro-civil rights demonstration)
(1) (2) (3) (4) (5) (6) (7)
Panel A. 2SLS
Change Black Share 0.053*** -0.003 -0.002 0.096*** 0.046 0.042 0.042(0.016) (0.023) (0.023) (0.014) (0.031) (0.036) (0.030)
Panel B. First Stage
Predicted Change Black 1.033*** 0.755*** 0.671*** 0.671***Share (0.094) (0.176) (0.171) (0.171)
1940 Black Share X X X X X1940 Dem Incumbent X X XWhite participants X
F-stat 119.5 18.39 15.43 15.43Observations 1,125 1,125 1,125 1,125 1,125 1,125 1,125
Notes: This table replicates Table 3 by aggregating the unit of analysis to the commuting zone (CZ).The dependent variable is the change in the probability of non-violent demonstrations in support of civilrights coordinated by the CORE. Columns 1 to 3 estimate equation (1) in the text with OLS, whileremaining columns report 2SLS estimates. The main regressor of interest is the change in the Blackshare, which is instrumented with the shift-share instrument described in equation (2) in the text fromcolumn 4 onwards. All regressions are weighed by 1940 CZ population, and control for state by periodfixed effects. 1940 Black share (resp 1940 Dem dummy) refers to interactions between period dummiesand the 1940 Black share (resp. a dummy equal to 1 if the Democratic vote share in 1940 was higherthan the Republicans vote share). Column 7 includes only those demonstrations that were joined by atleast some white participants. F-stat is the K-P F-stat for weak instruments. Robust standard errors,clustered at the CZ level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
64
Table A.11. Black in-Migration and White Population (Long Differences)
Dep. Variable Change White population
(1) (2) (3) (4)
Panel A. 2SLS
Change Black population 0.471 0.441 0.384 0.320
(0.412) (0.405) (0.383) (0.413)
Panel B. First Stage
Predicted Change Black population 2.504∗∗∗ 2.516∗∗∗ 2.526∗∗∗ 2.474∗∗∗
(0.365) (0.353) (0.339) (0.352)
F-stat 46.96 50.69 55.46 49.53
Observations 1,139 1,139 569 1,139
Baseline controls X X X
High urban X
Urban share X
Avg. change Black pop. 2,314 2,314 4,456 2,314
Avg. 1940 White pop. 71,001 71,001 120,557 71,001
Avg. change White pop. 12,058 12,058 19,860 12,058
Notes: The sample includes a panel of the 1,139 non-southern US counties (see Table A.1 for our definition ofsouthern states) for which electoral returns in Congressional elections are available for all Census years between1940 and 1970, and with at least one African American resident in 1940. The table estimates long differenceregressions, reporting 2SLS and first stage results in Panels A and B respectively. The dependent variable isthe 1940-1970 change in the white population in the county. The main regressor of interest is the correspondingchange in the Black population, instrumented with the shift-share instrument described in equation (2) in thetext. All regressions control for state fixed effects. Columns 2 to 4 further include i) the 1940 Black share;and ii) a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicans vote share).Column 3 restricts attention to counties with 1940 urban share of the population above the sample median(.356). Column 4 replicates column 2 by including the 1940 urban share of the population. F-stat is the KPF-stat for weak instruments. Robust standard errors, clustered at the county level, in parenthesis. Significancelevels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
65
Table A.12. Black in-Migration and white flight
Dep. variable: Change White Population Change White Populationin the county in central cities
(1) (2) (3) (4)
Panel A. 2SLS
Change Black population 0.155 0.236 -1.783*** -1.713***
(0.399) (0.347) (0.299) (0.293)
Panel B. First Stage
Predicted Change Black 2.784*** 2.701*** 1.722*** 1.876***
population (0.274) (0.266) (0.134) (0.203)
F-stat 102.9 103.1 166.3 85.40
Observations 351 351 154 154
Baseline Controls X X
Geography County County MSA MSA
Avg. Change Black pop. 18,661 18,661 42,919 42,919
Avg. 1940 White pop. 361,119 361,119 575,513 575,513
Avg. Change White pop. 59,198 59,198 -21,488 -21,488
Notes: In columns 1 and 2, the sample includes a panel of the 117 non-southern US counties contained in the 52metropolitan statistical areas (MSAs) included in Boustan (2010), for which electoral returns in Congressionalelections are available for all Census years between 1940 and 1970, and with at least one African Americanresident in 1940. Columns 3-4 focus on the 52 central cities contained in the 52 MSAs included in Boustan(2010). The dependent variable is the decadal change in the white population in the county (resp. in the centralcity) in columns 1-2 (resp. 3-4). The main regressor of interest is the change in the Black population in thecounty (resp. in the central city) in columns 1-2 (resp. 3-4), instrumented with the shift-share instrumentdescribed in equation (2) in the text. The table estimates stacked first difference regressions, reporting 2SLSand first stage results in Panels A and B, respectively. All regressions control for state by period fixed effects,and include interactions between period dummies and: i) the 1940 Black share; and ii) a dummy equal to 1if the Democratic vote share in 1940 was higher than the Republicans vote share. F-stat is the KP F-stat forweak instruments. Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗
p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
66
Table A.13. Black in-Migration and white flight (Long Differences)
Dep. variable: Change White Population Change White Populationin the county in central cities
(1) (2) (3) (4)
Panel A. 2SLS
Change Black population 0.313 0.365 -1.765*** -1.659***
(0.373) (0.320) (0.308) (0.278)
Panel B. First Stage
Change Predicted Black 2.872*** 2.762*** 5.249*** 5.596***
population (0.304) (0.300) (0.555) (0.584)
KP F-stat 71.52 66.51 102.7 89.23
Observations 115 115 52 52
Baseline Controls X X
Geography County County MSA MSA
Avg. Change Black pop. 18,661 18,661 42,919 42,919
Avg. 1940 White pop. 361,119 361,119 575,513 575,513
Avg. Change White pop. 59,198 59,198 -21,488 -21,488
Notes: In columns 1 and 2, the sample includes a panel of the 117 non-southern US counties contained in the 52metropolitan statistical areas (MSAs) included in Boustan (2010), for which electoral returns in Congressionalelections are available for all Census years between 1940 and 1970, and with at least one African Americanresident in 1940. Columns 3-4 focus on the 52 central cities contained in the 52 MSAs included in Boustan(2010). The dependent variable is the decadal change in the white population in the county (resp. in the centralcity) in columns 1-2 (resp. 3-4). The main regressor of interest is the change in the Black population in thecounty (resp. in the central city) in columns 1-2 (resp. 3-4), instrumented with the shift-share instrumentdescribed in equation (2) in the text. The table estimates long difference regressions, reporting 2SLS and firststage results in Panels A, and B, respectively. All regressions control for state fixed effects, and include: i)the 1940 Black share; and ii) a dummy equal to 1 if the Democratic vote share in 1940 was higher than theRepublicans vote share. F-stat is the KP F-stat for weak instruments. Robust standard errors, clustered at thecounty level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
67
Table A.14. Congressional Elections (CZ), Restricted Sample
Dep. Variable Change in
Democratic Vote Share Turnout
(1) (2) (3) (4)
Panel A. 2SLS
Change Black Share 2.759*** 2.704*** 0.673 0.942
(1.052) (0.920) (0.569) (0.617)
Panel B. First Stage
Predicted Change 0.671*** 0.763*** 0.671*** 0.763***
Black Share (0.171) (0.173) (0.171) (0.173)
Sample Baseline Restricted Baseline Restricted
(1960 Census) (1960 Census)
F-stat 15.43 19.44 15.43 19.44
Observations 1,125 375 1,125 375
Notes: The table replicates the CZ level results reported in Table A.9 by restricting the sample to CZs forwhich 1960 US Census data are available in columns 2 and 4 (columns 1 and 3 report results in column 6 ofTable A.9). The dependent variable is the change in the Democratic vote share in Congressional elections (resp.turnout) in columns 1-2 (resp. in columns 3-4). Panel B reports first stage coefficients. The main regressor ofinterest is the change in the Black share, which is instrumented with the shift-share instrument described inequation (2) in the text from column 4 onwards. All regressions are weighed by 1940 CZ population, and controlfor interactions between year dummies and: i) state dummies; ii) the 1940 Black share; and, iii) a dummy equalto 1 if the Democratic vote share in 1940 was higher than the Republicans vote share. F-stat is the K-P F-statfor weak instruments. Robust standard errors, clustered at the county level, in parenthesis. Significance levels:∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
68
Table A.15. Black in-Migration and Changes in Whites’ Characteristics
(1) (2) (3) (4) (5)
High Skilled In Manufacture In Labor Force Homeowner 65+
Panel A. 2SLS
Change Black Share -0.348 1.229 -0.389 -0.021 0.108
(0.626) (0.764) (0.424) (0.526) (0.334)
Panel B. First Stage
Predicted Change Black 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗
Share (0.179) (0.179) (0.179) (0.179) (0.179)
F-stat 26.39 26.39 26.39 26.39 26.39
Observations 125 125 125 125 125
1940 Mean 13.33 20.72 85.76 50.27 10.21
Avg. Change Black Share 3.372 3.372 3.372 3.372 3.372
Notes: The dependent variable is the 1940-1960 change in the share of white men above 18 not enrolled inschool who are: i) high skilled (column 1); ii) employed in manufacturing (column 2); iii) in the labor force(column 3); iv) homeowner (column 4); v) above the age of 65 (column 5). The table reports 2SLS results forthe 1940-1960 change in the Black share, instrumented with the shift-share IV described in the main text. Theanalysis is restricted to the 125 CZs for which demographic variables were available from the 1960 5% sampleof the micro-census. All regressions are weighed by 1940 population, control for state fixed effects, and includei) the 1940 Black share, and ii) a dummy equal to 1 if the Democratic vote share in 1940 was higher than theRepublicans vote share. F-stat is the KP F-stat for weak instruments. Robust standard errors, clustered at theCZ level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
69
Table A.16. Black Inflows and Whites’ Economic Outcomes
Dep. Variable 1940-1960 Change in
Labor Employed Manufacturing Log Occupational Log
Force Scores Wages
(1) (2) (3) (4) (5)
Panel A. 2SLS
Change Black Share -0.389 -0.512 1.229 0.003 0.077
(0.424) (0.488) (0.764) (0.007) (0.068)
Panel B. First Stage
Predicted Change 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗
Black Share (0.179) (0.179) (0.179) (0.179) (0.179)
F-stat 26.39 26.39 26.39 26.39 26.39
Observations 125 125 125 125 125
1940 Mean Dep. Var. 85.82 78.68 16.64 3.037 5.755
Notes: In columns 1 to 3, the dependent variable is the 1940-1960 change in the share of white men above18 not enrolled in school who are: i) in the labor force (column 1); ii) employed (column 2); iii) employedin manufacturing (column 3). In columns 4 and 5, the dependent variable is the 1940-1960 change in the logoccupational score and in log wages for white men above 18 not enrolled in school. The table reports 2SLSresults for the effects of the 1940-1960 change in the Black share, instrumented with the shift-share IV describedin the main text. The analysis is restricted to the 125 CZs for which demographic variables were availablefrom the 1960 5% sample of the micro-census. All regressions are weighted by 1940 population, and control forstate fixed effects, and include: the 1940 Black share, and the 1940 Democratic winner dummy. The bottomrow reports the 1940 mean of the dependent variable. F-stat is the K-P F-stat for weak instruments. Robuststandard errors, clustered at the CZ level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
70
Table A.17. Black Inflows and Whites’ Economic Outcomes: Unskilled and Manufacturing
Dep. Variable 1940-1960 Change in
Share Employed Log Wages Share Employed Log wages
(1) (2) (3) (4)
Panel A. 2SLS
Change Black Share -0.003 -0.004 -0.002 0.006
(0.004) (0.027) (0.003) (0.024)
Panel B. First Stage
Predicted Change 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗ 0.922∗∗∗
Black Share (0.179) (0.179) (0.179) (0.179)
F-stat 26.39 26.39 26.39 26.39
Observations 125 125 125 125
1940 Mean Dep. Var. 86.68 6.134 88.76 6.428
Sector Unskilled Unskilled Manufacturing Manufacturing
Notes: The dependent variable is the 1940 to 1960 change in the probability of employment and in log wages forwhite men not enrolled in school and above the age of 18 working in the unskilled sector (resp. in manufacturing)in columns 1 and 2 (resp. columns 3 and 4). The table reports 2SLS results for the effects of the 1940-1960change in the Black share, instrumented with the shift-share IV described in the main text. The analysis isrestricted to the 125 CZs for which demographic variables were available from the 1960 5% sample of the micro-census. All regressions are weighted by 1940 population, and control for state fixed effects, and include: the 1940Black share, and the 1940 Democratic winner dummy. The bottom row reports the 1940 mean of the dependentvariable. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the CZ level, inparenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
71
Table A.18. Placebo: Black in-Migration and White Movers (ANES)
Dep. Variable 1[Mover]
(1) (2) (3) (4)
Panel A. 2SLS
Change Black share -0.003 0.002 -0.023 -0.001
(0.018) (0.022) (0.035) (0.020)
Panel B. First Stage
Predicted change Black share 2.305∗∗∗ 2.180∗∗∗ 2.280∗∗∗ 2.262∗∗∗
(0.351) (0.321) (0.345) (0.336)
F-stat 43.20 46.22 43.57 45.20
Observations 3,911 1,598 1,562 1,307
Sample All MIP sample Democrats Republicans
Mean dep. variable 0.429 0.436 0.446 0.403
Notes: The sample is restricted to white ANES respondents living in the US North during the survey waves1956 to 1964. The dependent variable is a dummy equal to 1 if the individual is living in a state different fromhis/her state of birth. Column 2 restricts attention to respondents in the sample of Table 5, while columns 3and 4 consider self-identified Democrats and Republicans respectively. All regressions are weighed with ANESsurvey weights, include region fixed effects, and control for individual characteristics of respondents (gender, ageand education fixed effects, and marital status) as well as for 1940 state characteristics (Black share; Democraticincumbency in Congressional elections; share in manufacturing; share of workers in the CIO; urban share).F-stat refers to the K-P F-stat for weak instrument. Panel B reports the first stage. Robust standard errors,clustered at the state level, in parentheses. ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
72
Table A.19. Whites’ Voting Behavior and Party Identification (ANES)
Dep. Variable 1[Vote Democratic] 1[Identify Democratic]
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black Share 0.029∗∗∗ 0.038∗∗∗ 0.056∗∗ 0.028∗∗∗ 0.030∗ 0.048∗∗∗
(0.006) (0.007) (0.028) (0.007) (0.016) (0.011)
Panel B. First Stage
Predicted Change 2.317∗∗∗ 2.383∗∗∗ 2.041∗∗∗ 2.282∗∗∗ 2.382∗∗∗ 2.047∗∗∗
Black Share (0.374) (0.388) (0.278) (0.358) (0.378) (0.277)
F-stat 38.47 37.81 54.01 40.56 39.62 54.47
Observations 3,207 1,648 695 4,699 2,296 1,035
Sample All Non-movers 1964 All Non-movers 1964
Mean dep. variable .488 .489 .593 .412 .398 .441
Notes: The sample is restricted to white ANES respondents living in the US North for ANES survey waves1956 to 1964. In columns 1 to 3 the dependent variable is a dummy equal to 1 if the respondent voted (resp.intended to vote) for the Democratic Party in the previous (resp. upcoming) Congressional election. In columns4 to 6 the dependent variable is a dummy equal to 1 if the respondent identified with the Democratic Party.All regressions are weighed with ANES survey weights, include survey year and region fixed effects, and controlfor individual characteristics of respondents (gender, age and education fixed effects, and marital status) aswell as for 1940 state characteristics (Black share; Democratic incumbency in Congressional elections; share inmanufacturing; share of workers in the CIO; urban share). Columns 2 and 5 restrict the sample to respondentswho did not migrate across states during their life (as of the year of the interview). Columns 3 and 6 focus onsurvey wave 1964. F-stat refers to the K-P F-stat for weak instrument. Panel B reports the first stage. Robuststandard errors, clustered at the state level, in parentheses. ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
73
Table A.20. Additional Evidence on Whites’ Attitudes: Gallup
Dep. Variable 1[Object to half 1[Vote for 1[Approve Civil 1[Racial integration at the
pupils in school] Black candidate] Rights Act] right pace/not fast enough]
(1) (2) (3) (4)
Panel A. 2SLS
Change Black share -0.051*** 0.088*** 0.045*** 0.008
(0.019) (0.025) (0.013) (0.012)
Panel B. First Stage
Predicted change Black share 2.025*** 2.191*** 2.177*** 2.248***
(0.209) (0.352) (0.322) (0.302)
F-stat 93.71 38.79 45.67 55.30
Observations 851 2,073 931 17,478
Mean dep. variable 0.289 0.525 0.705 0.319
Notes: The sample is restricted to white Gallup respondents living in the US North and to years 1963-1964. The dependentvariable is a dummy equal to 1 if respondent: i) objects to having half of the classroom composed of Black pupils (column1); ii) would vote for a Black candidate, were her party nominating one (column 2); iii) approves the Civil Rights Actintroduced in 1964 (column 3); and iv) thinks that the process of racial integration is occurring at the right pace or not fastenough (column 4). All regressions include region and survey year fixed effects, and control for individual characteristics ofrespondents (gender and age and education fixed effects) as well as for 1940 state characteristics (Black share; Democraticincumbency in Congressional elections; share in manufacturing; share of workers in the CIO; urban share). F-stat refers tothe K-P F-stat for weak instrument. Panel B reports the first stage. Robust standard errors, clustered at the state level, inparentheses. ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
74
Table A.21. Heterogeneity by County Characteristics - Social and Cultural forces
Dep. Variable Change Pr.(Civil Rights Demonstration)
(1) (2) (3) (4) (5)
Panel A. Socially progressive counties
Change Black Share 0.240** 0.088*** 0.097** 0.048 0.148*
(0.106) (0.033) (0.043) (0.030) (0.083)
F-stat 6.48 4.43 4.29 4.05 2.70
Observations 1,683 2,089 1,709 1,699 1,701
Panel B. Socially conservative counties
Change Black Share 0.039** 0.025** 0.011 -0.017 0.005
(0.016) (0.010) (0.029) (0.020) (0.024)
F-stat 10.42 8.634 7.531 4.926 13.68
Observations 1,680 1,329 1,709 1,703 1,701
Proxy Discrimination Miscegenation Immigrant Young Southern
index laws share Whites born Whites
Notes: Socially progressive (resp. conservative) counties are defined as those with: the discrimination index from Qian andTabellini (2020) below (resp. above) the median in column 1, miscegenation laws not in place (resp. in place) in column 2,immigrant share of the population above (resp. below) the median in columns 3, share of the white population of age 35 orless above (resp. below) the median in column 4, share of whites born in a southern state below (resp. above) the medianin column 5. The table reports 2SLS results replicating the baseline specification (with interactions between year dummiesand: i) 1940 Black share; ii) 1940 Democratic incumbency dummy; iii) state fixed effects. All regressions are weighed by1940 population. Standard errors, clustered at the county level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05,∗ p< 0.1.
75
Table A.22. Heterogeneity by County Characteristics - Economic and political forces
Dep. Variable Change Pr.(Civil Rights Demonstration)
(1) (2) (3) (4) (5)
Panel A. Above median
Change Black Share 0.071*** 0.063*** 0.070*** 0.057*** 0.095**
(0.012) (0.024) (0.026) (0.015) (0.039)
F-stat 24.49 9.97 8.67 18.33 15.99
Observations 1,720 1,699 1,700 1,703 1,695
Panel B. Below median
Change Black Share -0.009 -0.025 -0.022 0.038*** 0.004
(0.040) (0.021) (0.026) (0.013) (0.039)
F-stat 3.66 6.89 7.61 6.74 11.75
Observations 1,694 1,703 1,702 1,715 1,696
1940 Characteristic Political Share in Share Share CIO Predicted
competition manufacturinge unskilled workers economic growth
Notes: Political competition in column 1 is defined as (one minus) the absolute value of the margin of victoryin 1940 Congressional elections. Predicted economic growth in column 6 is constructed by interacting the 1940industry shares in a county with the national growth in that industry over each of the subsequent three decades(and then summed across industries). The employment share in manufacturing and the share of unskilled workersrefer to white men in working age (15-64). The table reports 2SLS results replicating the baseline specification(with interactions between year dummies and: i) 1940 Black share; ii) 1940 Democratic incumbency dummy;iii) state fixed effects. All regressions are weighed by 1940 population. Standard errors, clustered at the countylevel, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
76
Table A.23. Probability that Civil Rights is Most Important Issue for Whites
Dep. Variable 1[Pro civil rights: Most important problem]
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black 0.107*** 0.002 0.032** -0.000 -0.020 0.025**
Share (0.020) (0.007) (0.015) (0.010) (0.013) (0.011)
Panel B. First Stage
Predicted Change 2.138*** 2.186*** 2.247*** 2.148*** 2.149*** 2.235***
Black Share (0.319) (0.316) (0.337) (0.309) (0.297) (0.334)
F-stat 44.95 47.95 44.36 48.41 52.28 44.66
Observations 465 1,133 670 930 517 1,083
Sample Union Non-Union Identified Identified Identified Identified
Members Members Democrat Non-Democrat Republican Non-Republican
Mean dep. variable 0.53 0.53 0.49 0.52 0.54 0.48
Notes: The sample is restricted to white ANES respondents living in the US North in years 1960 and 1964. The dependentvariable is a dummy equal to 1 if the respondent reports that supporting civil rights is among the most important issues facingthe country at the time of the interview (see online appendix D for exact wording and additional details on the constructionof the variable). The regressor of interest is the change in the Black share at the state level between 1940 and 1960, whichis instrumented with the predicted number of Black migrants over 1940 state population. All regressions include surveyyear and region fixed effects and individual controls of respondents (gender, age and education fixed effects and maritalstatus), and control for 1940 state characteristics (Black share; Democratic incumbency in Congressional elections; share inmanufacturing; share of workers in the CIO; urban share). Each column restricts attention to white respondents who belongto the group reported at the bottom of the table. F-stat is the K-P F-stat for weak instrument. Panel B reports the firststage. Robust standard errors, clustered at the state level, in parentheses. ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
77
Table A.24. Whites’ Feeling Thermometers
Dep. Variable Feeling Thermometer Towards
(1) (2) (3) (4)
Democrats Unions Blacks NAACP
Panel A. 2SLS
Change Black Share 2.646*** 2.924*** 0.843 1.329
(0.863) (0.855) (0.944) (0.903)
Panel B. First Stage
Predicted Change 2.045*** 2.041*** 2.041*** 2.031***
Black Share (0.276) (0.276) (0.276) (0.263)
F-stat 54.88 54.59 54.87 59.78
Observations 1,011 1,012 1,010 830
Mean Dep. Var. 68.70 57.64 60.82 53.55
Notes: : The sample is restricted to white ANES respondents living in the US North in 1964. The dependentvariable is the feeling thermometer towards each group at the top of the corresponding column. Higher valuesof the thermometer refer to warmer feelings. All regressions are weighed with ANES survey weights, includeregion fixed effects, and control for individual characteristics of respondents (gender, age and education fixedeffects, and marital status) as well as for 1940 state characteristics (Black share; Democratic incumbency inCongressional elections; share in manufacturing; share of workers in the CIO; urban share). F-stat refers to theK-P F-stat for weak instrument. Panel B reports the first stage. Robust standard errors, clustered at the statelevel, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
78
Table A.25. Change in Signatures on Discharge Petitions
Dep. Variable Change in Signatures on Discharge
Petitions per Legislator (1940s)
Total FEPC Anti-Lynching Poll-Tax
(1) (2) (3) (4)
Panel A. 2SLS
Change Black Share 1.011** 0.648** 0.197 0.168*
(0.417) (0.294) (0.133) (0.101)
Panel B. First stage
Predicted Change 0.940*** 0.940*** 0.940*** 0.940***
Black Share (0.350) (0.349) (0.350) (0.350)
F-stat 7.218 7.243 7.218 7.218
Observations 294 293 294 294
Baseline mean 1.742 0.357 0.338 1.046
dep. variable
Notes: The dependent variable is the change in number of signatures on discharge petition per legislatorbetween the beginning and the end of the 1940 decade (see the main text for more details). Column 1 considersall discharge petitions, while columns 2 to 4 focus on employment protection legislation (FEPC), Anti-Lynchinglegislation, and Poll Tax discharge petitions respectively. Data on discharge petitions were kindly shared byKathryn Pearson and Eric Schickler (see also Pearson and Schickler, 2009). Panel A reports 2SLS results, whilePanel B presents first stage estimates. All regressions are weighed by 1940 congressional district population,include state fixed effects, and control for: i) the 1940 Black share in the congressional district; ii) a dummy forDemocratic incumbency in the 78th Congress in the district; and iii) the ideology score in the district in the78th Congress. F-stat refers to the K-P F-stat for weak instrument. Robust standard errors, clustered at thecongressional district level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
79
Table A.26. Change in Signatures on Discharge Petitions, Restricted
Dep. Variable Change in Signatures on Discharge
Petitions per Legislator (1940s)
Total FEPC Anti-Lynching Poll-Tax
(1) (2) (3) (4)
Panel A. 2SLS
Change Black Share 1.181*** 0.695** 0.240* 0.192*
(0.453) (0.318) (0.132) (0.109)
Panel B. First stage
Predicted Change 1.006*** 1.006*** 1.006*** 1.006***
Black Share (0.379) (0.378) (0.379) (0.379)
F-stat 7.068 7.093 7.068 7.068
Observations 286 285 286 286
Baseline mean 1.747 0.362 0.338 1.047
dep. variable
Notes: The dependent variable is the change in number of signatures on discharge petition per legislator betweenthe beginning and the end of the 1940 decade (see the main text for more details). The sample is restrictedto congressional districts for which the change in the ideology score between the 78th and the 82nd Congressis available. Column 1 considers all discharge petitions, while columns 2 to 4 focus on EPL, Anti-Lynchinglegislation, and Poll Tax discharge petitions respectively. Data on discharge petitions were kindly shared byKathryn Pearson and Eric Schickler (see also Pearson and Schickler, 2009). Panel A reports 2SLS results, whilePanel B presents first stage estimates. All regressions are weighed by 1940 congressional district population,include state fixed effects, and control for: i) the 1940 Black share in the congressional district; ii) a dummy forDemocratic incumbency in the 78th Congress in the district; and iii) the ideology score in the district in the78th Congress. F-stat refers to the K-P F-stat for weak instrument. Robust standard errors, clustered at thecongressional district level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
80
Table A.27. Discharge Petitions (Levels on Changes)
Dep. Variable Number of petitions Share of petitions signed
per Legislator before 50th signature
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black Share 0.216* 0.412 0.138* 0.030** 0.053* 0.020*
(0.123) (0.269) (0.078) (0.013) (0.028) (0.012)
Panel B. First stage
Predicted Change 1.357*** 0.940*** 1.649*** 1.357*** 0.940*** 1.649***
Black Share (0.405) (0.326) (0.471) (0.405) (0.326) (0.471)
F-stat 11.24 7.218 10.69 11.24 7.218 10.69
Observations 588 294 294 588 294 294
Congress Period 78-82; 83-88 78-82 83-88 78-82; 83-88 78-82 83-88
Notes: The sample includes the 298 non-southern Congressional Districts that were representing non-southern US counties (seeTable A.1 for our definition of southern states) for which electoral returns in Congressional elections are available for all Censusyears between 1940 and 1970, with at least one African American resident in 1940, and for which data on signatures for dischargepetitions (Pearson and Schickler, 2009) were available. In columns 1-3, the dependent variable is the total number of signatureson discharge petitions per legislators during Congresses. In columns 4-6, the dependent variable is the share of discharge petitionssigned before the 50th signature. Columns 1 and 4 refer to both Congress periods 78-82 and 83-88, columns 2 and 5 to Congresses78-82 and columns 3 and 6 to Congresses 83-88. The main regressor of interest is the decadal change in the Black share in theCongressional District instrumented with the shift-share instrument described in the text. All regressions are weighed by 1940congressional district population, include state fixed effects, and control for: i) the 1940 Black share in the congressional district;ii) a dummy for Democratic incumbency in the 78th Congress in the district; and iii) the ideology score in the district in the 78thCongress. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the Congressional District level, inparenthesis Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
81
Table A.28. Black in-Migration and Political Polarization
Dep. Variable Change in ideology indicator of
elected Congress members
Liberal Moderate Moderate Conservative
Democrat Democrat Republican Republican
(1) (2) (3) (4)
Panel A. 78-82 Congresses
Change Black Share 0.331** -0.306* 0.165 -0.177**
(0.135) (0.181) (0.136) (0.082)
F-stat 7.068 7.068 7.068 7.068
Observations 286 286 286 286
Panel B. 82-88 Congresses
Change Black Share -0.020 0.013 -0.101* 0.108**
(0.042) (0.056) (0.057) (0.046)
F-stat 10.60 10.60 10.60 10.60
Observations 285 285 285 285
Notes: The dependent variable is the change in the ideology indicator of the Congress member in office. Theideology indicators are defined as: i) liberal (resp. moderate) Democrat if the legislator’s score was below(resp. above) the median score of the Democratic Party members in the 78th Congress; ii) moderate (resp.conservative) Republican if the legislator’s score was below (resp. above) the median score of the RepublicanParty members in the 78th Congress. Panel A refers to Congress period 78-82; Panel B refers to Congressperiod 82-88. All regressions are weighed by 1940 congressional district population and control for state by yearfixed effects and include interactions between year dummies and: i) the 1940 Black share in the congressionaldistrict; ii) a dummy for Democratic incumbency in the 78th Congress in the district; and iii) the ideology scorein the district in the 78th Congress. K-P F-stat for weak instruments. Robust standard errors, clustered at thecongressional district level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
82
B Appendix – Matching Counties to Time-Invariant Congres-sional Districts
When studying the effects of Black inflows on the behavior of northern legislators, we face two
main difficulties. First, while the African American population and other demographic variables
are measured at the county level, legislators’ behavior is available at the CD level. Second, the
boundaries of CDs change over time due to redistricting. We overcome both challenges by first
matching counties to CDs, and then by constructing a time-invariant cross-walk to map CDs
that get redistricted over time to their baseline geography.
B.1 County-CD Crosswalk
To overcome the first problem, and to assign to each CD the corresponding “Black in-migration
shock” we perform a spatial merge of 1940 county maps with CDs, following the procedure
used in Feigenbaum and Hall (2015).50 Since there is no one-to-one mapping between counties
and CDs, two cases can arise. First, some CDs are wholly contained within a single county; in
this case, we directly assign county level variables to CDs, assuming that the effect of Black
in-migration is uniform within the county. Second, some CDs straddle county boundaries. In
such cases, we assign county level values to the CD, weighting them by a county’s area share
of the CD.51 Figure B.1 displays the county (gray lines) to CD (Black lines) mapping just
described for the 78th Congress, restricting attention to non-southern states.
Figure B.1. CD-county Map
Notes: The figure presents a map of counties (gray lines) and Congressional Districts
(Black lines) for the non-South US during the 78th Congress.
50The only difference with their procedure is that we use counties rather than CZs.51Following Feigenbaum and Hall (2015), we test the robustness of our results using other weights, such as
maximum area.
83
B.2 Time Invariant CD Crosswalk
Until the early 1960s, there was no pre-determined rule mandating states to redraw CD bound-
aries after each decennial Census. Moreover, especially in the North, gerrymandering was
substantially less common than it is today (Snyder and Ansolabehere, 2008). Between 1900
and 1964, despite major demographic shifts induced by international and internal migration
(Boustan et al., 2014), redistricting across non-southern districts was typically non-strategic
(Engstrom, 2013). If anything, the lack of systematic redistricting rules likely introduced a pro-
rural bias: more densely populated areas (i.e. urban areas) grew gradually under-represented
at the CD level, likely diluting the effects of Black inflows, which were concentrated in urban
centers (see Figure 1 in the main text).52 However, even during the 1940-1965 period, the
boundaries of many CDs were changed, often multiple times. To overcome this empirical chal-
lenge, we develop a procedure that allows us to match all CDs between 1930 and 1970 to a
baseline Congress.53
We define the 78th Congress (January 6, 1943 to December 19, 1944) as our baseline
Congress year for two main reasons. First, although the 76th Congress might have been a
more natural choice (as it corresponds to the 1940 Census year), several CDs underwent re-
districting between this Congress year and the 78th Congress. In contrast, very few states
redistricted between the 78th and the 82nd Congress. Second, Congress 78th is the earliest
Congress for which CD-level population estimates are available from Adler (2003), thus allow-
ing us to benchmark the population figures estimated in our procedure with other measures.
We thus rely on Congress 78 as our baseline year, and consider the following two Congress
periods: 78 to 82, which we match to the 1940 to 1950 Census decade; and, 83 to 88, which
we match to the 1950 to 1960 year.54 We perform a number of robustness checks to show
that our results do not depend on the choice of the baseline Congress year, and that they are
qualitatively similar when restricting the sample to CDs that did not undergo redistricting over
the 78 to 82 Congress period.
Using this timing convention, for every Congress between 71 and 91, we perform a spatial
merge between CD maps and the map corresponding to the 78th Congress. Then, political
outcomes (e.g. ideology scores, number of discharge petitions signed by legislators, etc.) are
collapsed to the 78th Congress using a weighting procedure similar to that adopted when
matching counties to CDs. The logic of our strategy is simple: we fix the 1944 (i.e. the 78th
Congress) geography of CDs, and we link them to CDs that represented the same geographic
area in subsequent (or previous) Congress years.55 Then, we calculate a weighted average of
political outcomes that correspond to the area originally represented by CDs according to the
1944 map.
To illustrate this procedure, we ask how the 78th Congress would have looked like, had
52This observation suggests that our analysis should identify a lower bound for the effects of Black inflows onlegislators’ (pro-civil rights) behavior.
53While our analysis focuses on years after 1940, we also construct the cross-walk for the pre-1940 decade inorder to perform several robustness and falsification checks.
54The reason to consider the 88th Congress in the second decade is that this was the Congress that approvedthe CRA.
55When states have more than one district, we drop at-large Congressional seats from the spatial merge (e.g.at-large seats for the state of New York are dropped between 1933 and 1945).
84
its geography persisted until Congress 86. We now explain how we proceed to collapse the
political outcomes corresponding to the geography of Congress 86 “back” to that of Congress
78. Suppose that the area represented by a single CD in Congress 78 gets split in two separate
CDs by Congress 86. To assign political variables of new CDs back to the level of the original
CD, we adopt a weighting procedure, based on weights constructed in four steps. First, we
overlay the map of the initial CD to that of the two CDs in Congress 86, and divide the area
in cells derived by this spatial merge. Second, we assign the 1940 county population to each
cell in proportion to the area share of the cell that is included in the county. Third, we sum
over all cells that compose the CD to obtain an estimate of CD population as of Congress 78.
Finally, we divide the area of each cell by such estimated CD population.
Political variables corresponding to the geography of the 78th Congress for subsequent
Congress years are computed by taking the weighted average of the outcomes of the newly
formed CDs, using the weights constructed as explained above. In Appendix C, we validate
the accuracy of this approach by replicating our (baseline) county-level results for the Demo-
cratic vote share using CD level data from Swift et al. (2000). Reassuringly, when conducting
the analysis at the CD level, results remain qualitatively and quantitatively similar to those
reported in the main text (see Table 2).
85
C Appendix – Robustness Checks
In this section we present a variety of robustness checks. First, we show that the instrument
is uncorrelated with county-specific “pull factors” that might have influenced pre-1940 Black
settlements, and verify that results are unchanged when interacting year dummies with a
variety of 1940 county characteristics. Second, we document that our findings are not driven
either by pre-existing trends or by the simultaneous inflow of southern born white migrants.
Third, we show that results are robust to omitting potential outliers, considering alternative
proxies for support for the Democratic Party, and estimating different specifications. Fourth,
we construct an alternative version of the instrument that predicts Black out-migration from
each southern state exploiting only variation across local push factors. Finally, we document
that CD-level results: i) are robust to using different timing conventions to map Black inflows
to Congress periods; ii) are not influenced by pre-existing trends; and, iii) are unlikely to be
driven by strategic gerrymandering.
C.1 Initial Shares, County Characteristics, and Local Shocks
In Table C.1, we start by investigating if the instrument constructed in equation (2) in the
main text is correlated with county-specific pull factors, such as WWII contracts, New Deal
spending, and the vote share of Franklin Delano Roosevelt (FDR) in 1932 Presidential elections.
As discussed in Boustan (2016), the surge in demand across northern and western factories
triggered by WWII was one of the pull factors of the Great Migration. Similarly, the generosity
of New Deal spending and local support for FDR might have influenced the location decision of
African Americans prior to 1940, while at the same time having long-lasting effects on political
conditions across northern counties.
The dependent variable in Table C.1 is the change in predicted Black in-migration, scaled
by 1940 county population. The main regressor of interest is one of the three variables described
above – WWII spending per capita (Panel A), generosity of New Deal (Panel B), and 1932
FDR vote share (Panel C). Columns 1 to 3 consider each decade separately, whereas column 4
focuses on the long difference (1940-1970) change in predicted Black in-migration. We always
include the set of controls used in our most preferred specification – i.e., state dummies, the
1940 Black share, and a dummy equal to 1 if in 1940 the Democratic vote share was higher than
the Republican vote share in Congressional elections – and weigh regressions by 1940 county
population. Reassuringly, in all cases the coefficient is imprecisely estimated and quantitatively
small.
86
Table C.1. Placebo Checks
Dep. Variable Predicted Change in Black Share (Mean: 0.916)
(1) (2) (3) (4)
Panel A. WWII (Mean: 0.648)
Spending per capita 0.041 0.051 0.043 0.092
(0.035) (0.043) (0.037) (0.107)
Observations 1,139 1,139 1,140 1,139
Panel B. New Deal (Mean: 0.143)
Spending per capita -0.284 -0.268 -0.148 -0.514
(0.183) (0.225) (0.192) (0.525)
Observations 1,139 1,139 1,140 1,139
Panel C. Vote Share FDR (Mean: 53.93)
1932 Vote Share FDR -0.005 -0.004 -0.003 -0.009
(0.004) (0.004) (0.003) (0.010)
Observations 1,117 1,115 1,116 1,117
Decade 1940-1950 1950-1960 1960-1970 1940-1970
Notes: The dependent variable is the change in the predicted number of Black migrants over 1940 county
population. Each column considers the period specific to the decade reported at the bottom of the table. All
regressions are weighed by 1940 county population, and control for state dummies, for the 1940 Black share, and
for a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicans vote share. In
Panels A, B, and C the main regressor of interest is WWII spending per capita, New Deal spending per capita,
and the vote share for FDR in the 1932 Presidential elections. Robust standard errors, clustered at the county
level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
Next, to address concerns that 1940 Black settlements (from each southern state) might
be correlated with county-specific characteristics that may have had a time varying effect on
changes in political conditions, we interact period dummies with several 1940 county charac-
teristics. Column 1 of Table C.2 and Table C.3 replicates the baseline specification estimated
in Tables 2 and 3 (column 6) in the main text. For completeness, we also report first stage
estimates at the bottom of each table. Columns 2 to 6 augment the baseline specification by
including interactions between period dummies and, respectively, the 1940: i) urban share; ii)
share of employment in manufacturing; iii) male employment to population ratio; iv) fraction
of immigrants; and, v) county population. Reassuringly, the coefficient remains stable and, for
both the Democratic vote share and CORE demonstrations, highly statistically significant.56
In column 7, we augment the baseline specification by separately controlling for a predicted
measure of labor demand growth constructed using a Bartik-type approach (similar to e.g.
Sequeira et al., 2020, and Tabellini, 2020). Restricting attention to non-southern counties, we
56In columns 5 and 6, the KP F-stat falls below conventional levels, due to the stringent nature of the exerciseperformed.
87
first compute the 1940 share of employment in each 1-digit industry in each county; then, we
interact these initial shares with the national growth rate of employment in that industry.57
Once again, results are quantitatively similar to, in fact slightly larger than, those reported in
column 1.
Finally, we deal with the possibility that the 1940 share of Blacks from each southern
state were not independent of cross-county pull factors systematically related to settlers’ state
of origin (Goldsmith-Pinkham et al., 2020). To do so, we replicate our county-level results
by interacting year dummies with the share of Blacks from each southern state, i.e. shsc in
equation (2) in the main text. In Figures C.1, C.2, and C.3, we plot 2SLS coefficients for the
effects of changes in the Black share onthe change in the Democratic vote share, in turnout, and
in CORE demonstrations respectively. The very first dot on the left of the graphs represents the
coefficient from our baseline specification (see also column 6 in Tables 2 and 3). Reassuringly,
both the precision and the magnitude of our estimates are stable across specifications.
57To more precisely proxy for labor demand shocks in non-southern industries, we compute the nationalgrowth rate for the non-South only. Results are unchanged when including the US South to compute nationaldemand growth.
88
Table C.2. Congressional Elections: Controlling for 1940 County Characteristics
(1) (2) (3) (4) (5) (6) (7)
2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS
Panel A: Democratic Vote Share
Change Black Share 1.650*** 2.337*** 1.673*** 1.636*** 1.700*** 2.304*** 1.920***
(0.286) (0.555) (0.363) (0.279) (0.456) (0.718) (0.417)
Panel B. Turnout
Change Black Share 0.390* 0.795* 0.381 0.394* 0.746 0.874 0.286
(0.235) (0.438) (0.268) (0.235) (0.466) (0.653) (0.280)
Panel C. First Stage
Change Predicted 1.148*** 0.741*** 0.994*** 1.153*** 0.806*** 0.710** 0.999***
Black share (0.311) (0.240) (0.300) (0.311) (0.308) (0.322) (0.312)
Control Baseline Urban Manufacturing Employment Immigrant County Predicted
Share Share to population share population demand growth
F-stat 13.65 9.523 10.96 13.76 6.825 4.857 10.26
Observations 3,418 3,418 3,418 3,418 3,418 3,418 3,391
Notes: The table replicates the main specification (which is also reported in column 1) for results reported in Table 2 (column 6)by including the interaction between period dummies and, respectively, the 1940: i) urban share (column 2); ii) employment share inmanufacturing (column 3); iii) male employment to population ratio (column 4); iv) immigrant share (column 5); v) county population.In column 7, the baseline specification is augmented by separately controlling for a measure of predicted industrial growth constructedwith a Bartik-style strategy described in the text of the Appendix. Panel C reports the first stage for the 2SLS results presented in PanelsA and B. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the county level, in parenthesis. Significancelevels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
89
Table C.3. CORE Demonstrations: Controlling for 1940 County Characteristics
Dep. variable Change in Pr.(Pro-civil rights demonstration)
(1) (2) (3) (4) (5) (6) (7)
Panel A. Democratic Vote Share
Change Black Share 0.047*** 0.046*** 0.054*** 0.047*** 0.053*** 0.037** 0.032**
(0.011) (0.017) (0.013) (0.011) (0.018) (0.017) (0.014)
Panel B. First Stage
Predicted Change 1.148*** 0.741*** 0.994*** 1.153*** 0.806*** 0.710** 0.999***
Black share (0.311) (0.240) (0.300) (0.311) (0.308) (0.322) (0.312)
Control Baseline Urban Manufacturing Employment Immigrant County Predicted
Share Share to population share population demand growth
F-stat 13.65 9.523 10.96 13.76 6.825 4.857 10.26
Observations 3,418 3,418 3,418 3,418 3,418 3,418 3,391
Notes: The table replicates the main specification (which is also reported in column 1) for results reported in Table 3 (column 6)by including the interaction between period dummies and, respectively, the 1940: i) urban share (column 2); ii) employment share inmanufacturing (column 3); iii) male employment to population ratio (column 4); iv) immigrant share (column 5); v) county population.In column 7, the baseline specification is augmented by separately controlling for a measure of predicted industrial growth constructedwith a Bartik-style strategy described in the text of the Appendix. Panel B reports the first stage for the 2SLS results presented in PanelA. F-stat is the K-P F-stat for weak instruments. Robust standard errors, clustered at the county level, in parenthesis. Significance levels:∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
90
Figure C.1. Interacting Year Dummies with Initial Shares: Democratic Vote Share
Notes: The figure plots the 2SLS point estimate (with corresponding 95% confidence
intervals) for the effects of a change in the Black share on the Democratic vote share,
augmenting the baseline specification reported in Table 2 with interactions between
period dummies and the 1940 share of Blacks born in each southern state. The very
first dot on the left reports the coefficient for the baseline specification.
91
Figure C.2. Interacting Year Dummies with Initial Shares: Turnout
Notes: The figure plots the 2SLS point estimate (with corresponding 95% confidence
intervals) for the effects of a change in the Black share on turnout, augmenting the baseline
specification reported in Table 2 with interactions between period dummies and the 1940
share of Blacks born in each southern state. The very first dot on the left reports the
coefficient for the baseline specification.
92
Figure C.3. Interacting Year Dummies with Initial Shares: CORE Demonstrations
Notes: The figure plots the 2SLS point estimate (with corresponding 95% confidence
intervals) for the effects of a change in the Black share on turnout, augmenting the baseline
specification reported in Table 3 with interactions between period dummies and the 1940
share of Blacks born in each southern state. The very first dot on the left reports the
coefficient for the baseline specification.
C.2 Testing for Pre-Trends
In Table C.4, we perform a key placebo check, and show that there is no correlation between pre-
period changes in the outcomes of interest and the (instrumented) change in the Black share.
Since 1942 is the first year in which CORE demonstrations occurred, we conduct this exercise
only for the Democratic vote share and for turnout in Congressional elections.58 Table C.4
reports results for the Democratic vote share (resp. turnout) in columns 1 to 3 (resp. 4 to 6).
To ease comparisons, columns 1 and 4 present the baseline specification (Table 2, column 6);
next, in columns 2 and 5, we replicate our analysis restricting attention to counties for which
“pre-trends” regressions can be estimated. As it appears, while results for the Democratic
vote share become slightly larger, those for turnout double in size and become more precisely
estimated.
Finally, in columns 3 and 6, we turn to the formal test for pre-trends, regressing the 1934 to
1940 change in the Democratic vote share and in turnout against the 1940 to 1970 instrumented
change in the Black share.59 When constructing the “pre-1940” change in political outcomes, we
consider the first election year after 1932 in order to make sure that results are not confounded
by post-New Deal realignment (Caughey et al., 2020; Schickler, 2016). However, our findings
are unchanged when using other election years, such as 1930 or 1932. Focusing on “pre-
58Appendix C.5 below conducts a similar test (at the CD level) for legislators’ ideology.59As noted in the main text, the instrument is scaled by 1940 population so as not to contaminate it with the
potentially endogenous contemporaneous population (Card and Peri, 2016).
93
trends” regressions, reassuringly, the coefficient is not statistically significant and, especially
for the Democratic vote share (column 3), very different from that estimated in the baseline
specification.60
Table C.4. Testing for Pre-Trends: Congressional Elections
Dep. Variable Change Democratic vote share Change Turnout
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black 1.650*** 1.954*** -0.400 0.390* 0.809** 0.312
Share (0.286) (0.453) (0.370) (0.235) (0.357) (0.212)
Panel B. First Stage
Change predicted 1.148*** 0.738*** 0.778*** 1.148*** 0.738*** 0.778***
Black share (0.311) (0.229) (0.245) (0.311) (0.229) (0.245)
F-stat 13.65 10.39 10.13 13.65 10.39 10.13
Observations 3,418 3,401 1,138 3,418 3,401 1,138
Specification Baseline Restricted Pre-Trends Baseline Restricted Pre-Trends
Notes: Panel A reports 2SLS estimates for the change in the Democratic vote share (resp. turnout) in Con-
gressional elections in columns 1-3 (resp. 4-6). Columns 1 and 4 report the baseline specification (see Table 2,
column 6), and columns 2 and 5 replicate the baseline specification restricting attention to counties for which
the change in the Democratic vote share and turnout between 1934 and 1940 can be computed. Columns 3
and 6 estimate first difference regressions for the 1934-1940 change in the Democratic vote share and in turnout
against the 1940 to 1970 instrumented change in the Black share. The pre-period is defined using the first
election year after the New Deal election of 1932, i.e. 1934. However, results are unchanged when using other
timing conventions. All regressions are weighed by 1940 county population and include state dummies, the 1940
Black share, and a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicans vote
share. These variables are interacted with year dummies in columns 1-2 and 4-5 (as in the main text). F-stat
is the K-P F-stat for weak instruments. Robust standard errors, clustered at the county level, in parenthesis.
Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
C.3 Additional Robustness Checks
C.3.1 Alternative Specifications
In our baseline analysis, we interact year dummies with a dummy equal to 1 if the 1940
Democratic vote share in Congressional elections were greater than the Republican one to allow
counties to be on different trends depending on Democratic incumbency (and potentially deal
with mean reversion). To more flexibly account for initial support for the Democratic Party, in
column 2 of Tables C.5 and C.6, we replicate the baseline analysis (reported in column 1 to ease
comparisons) by interacting the 1940 Democratic vote share with year dummies. Reassuringly,
60For turnout, the point estimate is close to that of the baseline specification for the full sample. Yet, oncecompared to the baseline estimates obtained for the “restricted” sample (column 5), the coefficient is almostone third smaller.
94
all results remain in positive and statistically significant. If anything, they become larger, and
more precisely estimated, especially for turnout (Panel B of Table C.5).61
Next, in column 3 of Tables C.5 and C.6, we replicate the baseline specification by estimating
unweighted regressions. Reassuringly, even though results for turnout are no longer statistically
significant, they remain strong and precisely estimated for both the Democratic vote share and
CORE demonstrations. In this case, the KP F-stat for weak instruments falls slightly below
conventional levels. But, the main message remains unchanged.
C.3.2 Dropping Potential Outliers
As discussed in the main text, some areas of the US North and West, such as Chicago, Detroit,
and Los Angeles, received a disproportionately large inflow of Black migrants between 1940
and 1970. Others, instead, received very few African Americans. In our main analysis we omit
counties with zero Black individuals in 1940, so as to compare counties that received different
numbers of migrants with each other (and exclude from this comparison counties that did not
have any Black resident in 1940). We now show that all results are robust to restricting the
sample in different ways.
First, in columns 4 and 5 of Tables C.5 and C.6, we restrict attention to counties for which
the predicted and the actual Black share was strictly positive in all decades between 1940 and
1970. Not surprisingly, results are unchanged. Next, in column 6 (resp. 7), we exclude counties
at the top 1st (resp. 5th) and at the bottom 99th (resp. 95th) percentiles of the distribution
of changes in Black migration. Also in this case, results remain in line with – in fact stronger
than – those of our baseline specification.
C.3.3 Controlling for Southern White In-Migration
Yet another potential concern is that Black in-migration might be correlated with simultaneous
white inflows from the South. As documented in Gregory (2006) among others, between 1940
and 1970 even more whites than Blacks left the US South. The historical evidence suggests that
African Americans were significantly more likely than whites to settle in metropolitan areas
either in the Northeast or in the West, while white migration was more evenly distributed
across the non-South (Gregory, 1995). However, it is still possible that the patterns of white
and Black migration from the South were correlated with each other. If this were to be the
case, at least part of our findings might be due to the arrival of white – rather than Black –
migrants. Due to data limitations, we cannot measure the actual change in southern born white
migrants after 1940 at the county level. However, to at least partly overcome this problem,
we construct a predicted measure of white in-migration from the US South implementing the
same procedure used to construct the instrument for Black in-migration (see equation (2) in
the main text).
Specifically, we first compute the share of whites born in each southern state who were
living in a non-southern county as of 1940. Next, we interact these shares with the number
61Reassuringly, the first stage, reported in the bottom panel of both tables, remains strong and the KP F-statfor weak instruments above conventional levels.
95
Table C.5. Additional Robustness Checks: Congressional Elections
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. Democrat Vote Share
Change Black Share 1.650*** 2.095*** 1.936*** 1.649*** 1.650*** 1.757*** 2.662*** 1.534***
(0.286) (0.424) (0.555) (0.286) (0.288) (0.377) (0.651) (0.243)
Panel B. Turnout
Change Black Share 0.390* 0.690** 0.198 0.392* 0.394* 0.515* 0.700* 0.356*
(0.235) (0.301) (0.328) (0.236) (0.234) (0.292) (0.383) (0.215)
Panel C. First Stage
Predicted Change 1.148*** 0.808*** 0.392*** 1.147*** 1.146*** 0.981*** 0.762*** 1.250***
Black Share (0.311) (0.232) (0.137) (0.311) (0.311) (0.286) (0.246) (0.290)
F-stat 13.65 12.11 8.199 13.61 13.58 11.80 9.597 18.62
Specification Baseline 1940 Dem vote Unweighted Drop IV Drop Black Trim top 99 and Trim top 95 and Southern White
share equal to 0 share equal to 0 bottom 1 pctile bottom 5 pctile in-Migration
Observations 3,418 3,414 3,418 3,295 3,251 3,342 3,081 3,418
Notes: The table replicates the main specification (which is also reported in column 1) for results reported in Table 2 (column 6) by: i) replacing the interaction between perioddummies and the 1940 Democratic incumbency dummy with that with the 1940 Democratic vote share in Congressional elections (column 2); ii) estimating unweighted regressions(column 3); iii) considering only counties with predicted (resp. actual) Black share strictly positive in all decades in column 4 (resp. column 5); iv) trimming counties at the top 1st
(resp. 5th) and at the bottom 99th (resp. 95th) percentiles of the distribution of changes in Black migration in column 6 (resp. column 7); and v) controlling for predicted southernwhite in-migration (column 8). Panel C reports the first stage for the 2SLS results presented in Panels A and B. F-stat is the K-P F-stat for weak instruments. Robust standarderrors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ** p< 0.05,* p< 0.1.
96
Table C.6. Additional Robustness Checks: CORE Demonstrations
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A.2SLS
Change Black Share 0.047*** 0.056*** 0.032** 0.047*** 0.047*** 0.068*** 0.051** 0.046***
(0.011) (0.016) (0.013) (0.011) (0.011) (0.018) (0.021) (0.010)
Panel B. First Stage
Predicted Change 1.148*** 0.808*** 0.392*** 1.147*** 1.146*** 0.981*** 0.762*** 1.250***
Black Share (0.311) (0.232) (0.137) (0.311) (0.311) (0.286) (0.246) (0.290)
F-stat 13.65 12.11 8.199 13.61 13.58 11.80 9.597 18.62
Specification Baseline 1940 Dem vote Unweighted Drop IV Drop Black Trim top 99 and Trim top 95 and Southern White
share equal to 0 share equal to 0 bottom 1 pctile bottom 5 pctile in-Migration
Observations 3,418 3,414 3,418 3,295 3,251 3,342 3,081 3,418
Notes: The table replicates the main specification (which is also reported in column 1) for results reported in Table 3 (column 6) by: i) replacing the interaction betweenperiod dummies and the 1940 Democratic incumbency dummy with that with the 1940 Democratic vote share in Congressional elections (column 2); ii) estimating unweightedregressions (column 3); iii) considering only counties with predicted (resp. actual) Black share strictly positive in all decades in column 4 (resp. column 5); iv) trimmingcounties at the top 1st (resp. 5th) and at the bottom 99th (resp. 95th) percentiles of the distribution of changes in Black migration in column 6 (resp. column 7); and v)controlling for predicted southern white in-migration (column 8). Panel B reports the first stage for the 2SLS results presented in Panel A. F-stat is the K-P F-stat for weakinstruments. Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ** p< 0.05,* p< 0.1.
97
of white migrants from each southern state in each decade between 1940 and 1970. Finally,
for each non-southern county and for each decade, we sum the predicted number of whites
moving from each origin over all southern states to obtain the total number of (predicted)
white migrants moving to county c during decade τ . In formulas, this measure is given by:
ZWcτ =∑
j∈SouthshwjcWhjτ (3)
where shwjc is the share of whites born in southern state j and living in non-southern county c
in 1940, relative to all whites born in j living outside this state; and Whjτ is the number of
whites who left southern state j during decade τ .
Then, in column 8 of Tables C.5 and C.6, we augment our baseline specification by sep-
arately controlling for the predicted southern white in-migration. Reassuringly, in all cases,
results are barely affected.
C.3.4 Additional Outcomes
In the paper, we focus on the Democratic vote share as the main electoral outcome of interest.
In Table C.7, we verify that results are unchanged when considering different proxies for support
for the Democratic Party in Congressional elections. Column 1 presents our main 2SLS results
for the Democratic vote share (Table 2, column 6). Next in columns 2 and 3, the dependent
variable is defined respectively as the Democratic vote margin and as a dummy equal to 1 if
the Democratic vote share was larger than the Republicans vote share. In both cases, Black
in-migration is associated with an increase in support for the Democratic Party.
98
Table C.7. Additional Outcomes: Congressional Elections
Dep. variable Democratic Democrats-Republicans 1[Higher Democratic
Vote Share Vote Margin Vote Share]
(1) (2) (3)
Panel A.2SLS
Change Black Share 1.650*** 3.184*** 0.031***
(0.286) (0.555) (0.010)
Panel B. First Stage
Predicted Change 1.148*** 1.147*** 1.152***
Black Share (0.311) (0.309) (0.312)
F-stat 13.65 13.79 13.62
Observations 3,418 3,401 3,325
Notes: Panel A presents 2SLS results. Column 1 replicates the baseline specification for the effects of changes
in the Black share on the Democratic vote share (Table 2, column 6). In columns 2 and 3, the dependent variable
is, respectively, the Democrats-Republicans vote margin in Congressional elections, and a dummy equal to 1
if the Democratic vote share was higher than the Republicans vote share in Congressional elections. Panel B
reports the first stage. All regressions are weighed by 1940 county population and include interactions between
year dummies and: i) state dummies; ii) the 1940 Black share; and iii) a dummy equal to 1 if the Democratic
vote share in 1940 was higher than the Republicans vote share.F-stat is the K-P F-stat for weak instruments.
Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ** p< 0.05,
* p< 0.1.
C.3.5 Stacked Panel Specification
Finally, we verify that our results are robust to estimating stacked panel regressions separately
controlling for county fixed effects, rather than taking the model in (stacked) first differences.
Specifically, we stack the data for the four decades between 1940 to 1970 (included), and run
a regression of the form:
yct = ξc + δst + βBlct + γXct + uct (4)
where yct refers to the Democratic vote share and turnout in Congressional elections or to the
probability of CORE demonstrations in county c in year t, ξc and δst are county and state by
year fixed effects, and Blct is the Black share in county c in year t. As for the stacked first
difference specification, Xct includes interactions between year dummies and baseline: i) Black
share; and ii) Democratic incumbency in Congressional elections.62
In our baseline specification, we used predicted Black inflows in each decade to instrument
for the change in Black population. However, when estimating equation (4), an instrument is
needed for Black population in each year from 1940 to 1970. That is, 1940 can no longer be
used as “baseline” year to predict Black inflows. Also, since we are now interested in Black
population (relative county population) rather than in its change, we need an instrument for the
62Since in a stacked panel setting 1940 is our first estimation year, we measure the baseline Black share in1930 and the Democratic incumbency in 1934. Results are unchanged if we measure both variables in 1940.
99
stock – and not the change – of Blacks in the county. We thus modify the baseline instrument
constructed in the main text in two ways. First, we use 1930 settlements of African Americans
across northern counties to apportion post-1930 outmigration from the South. Second, after
predicting the inflow of Blacks to county c in the ten years prior to year t, we recursively add
previous predicted inflows to generate a measure of predicted stock.63
With this instrument at hand, we proceed to estimate equation (4) with 2SLS. We report
results in Panel A of Table C.8, presenting first stage in Panel B. Focusing on Democratic vote
share, turnout, and CORE demonstrations respectively, we report the baseline (stacked first
difference) specification in columns 1-3-5 to ease comparisons, and the stacked panel regressions
in columns 2-4-6. The KP F-stat for weak instrument falls below conventional levels when
estimating the stacked panel specifications. However, the main take-away is unchanged. Black
in-migration has a positive and statistically significant effect on the Democratic vote share and
on the frequency of CORE demonstrations, and a small and imprecisely estimated effect on
turnout.
63As before, we scale the predicted number of Blacks by 1940 county population. Results are unchanged whendividing it by 1930 population.
100
Table C.8. Stacked Panel Specification
Dep. variable Democratic Vote Share Turnout 1[CORE Demonstrations]
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black Share 1.650*** 1.394*** 0.390* 0.128 0.047*** 0.045***
(0.286) (0.413) (0.235) (0.172) (0.011) (0.015)
Panel B. First stage
Predicted Change 1.148*** 1.118*** 1.148*** 1.118*** 1.148*** 1.118***
Black Share (0.311) (0.405) (0.311) (0.405) (0.311) (0.405)
F-stat 13.65 7.438 13.65 7.438 13.65 7.438
Observations 3,418 4,328 3,418 4,328 3,418 4,328
Specification Stacked first Stacked panel Stacked first Stacked panel Stacked first Stacked panel
differences diferences differences
Notes: : The table replicates the baseline stacked first difference specification using a stacked panel specification. The dependent
variable is the (resp. the change in) Democratic vote share, turnout, and probability of CORE demonstrations in columns 2-4-6
(resp. in columns 1-3-5). Panel A reports 2SLS estimates, and Panel B presents the first stage. All regressions are weighed by 1940
county population. Columns 1-3-6 include interactions between year dummies and: i) state dummies; ii) the 1940 Black share; and,
iii) a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicans vote share. Columns 2-4-6 include
county and state by year fixed effects, and control for interactions between year dummies and: i) the 1930 Black share; and, ii) a
dummy equal to 1 if the Democratic vote share in 1934 was higher than the Republicans vote share. Results in columns 2-4-6 are
unchanged when including interactions using 1940 (rather than pre-1940 values). F-stat is the K-P F-stat for weak instruments.
Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
101
C.4 Push Factors Instrument
C.4.1 Instrument Construction and Zeroth Stage
Borusyak et al. (2018) note that the validity of shift-share designs can be guaranteed if the
“shifts” – in our case, decadal Black migration from each southern state – are exogenous to
local conditions. They propose a correction method, where the “shift-share” instrument is
expressed in terms of the “shift” components. This method, however, can be implemented
only when the number of “shifts” is large. Unfortunately, in our setting, we can only rely on 15
southern states, and so we cannot directly implement the transformation proposed in Borusyak
et al. (2018).
Nevertheless, we provide evidence that southern (state) migration flows are orthogonal to
northern (county) conditions. We construct a modified version of the instrument that, rather
than using actual Black out-migration, estimates it exploiting variation solely induced by local
push factors. Following Boustan (2010, 2016) and Derenoncourt (2018), we model emigration
from each southern county for each decade between 1940 and 1970 as a function of local push
factors. In particular, we estimate an equation of the form
migkjτ = αj + βτPushkjt0 + ukjτ (5)
where migkjτ is the Black net migration rate in county k of southern state j during decade τ ,
and Pushkjt0 is a vector of economic and political conditions at baseline, which we allow to have
a time-varying effect across decades. These include the 1940: share of land cultivated in cotton;
share of farms operated by tenants; share of the labor force in, respectively, manufacturing,
mining, and agriculture. As in Boustan (2016), we also include WWII spending per capita and
the 1948 vote share of Strom Thurmond in Presidential elections.64
Our most preferred specification includes state fixed effects, αj , but results are unchanged
when omitting them (see also Boustan, 2016). Finally, in contrast with Boustan (2010, 2016),
we fix the characteristics of southern counties to 1940 (or, for Thurmond vote share, 1948)
rather than using the beginning of each decade to reduce concerns of correlated shocks be-
tween northern and southern counties.65 As an additional robustness check, we also selected
the southern county characteristics to predict Black out-migration using the Least Absolute
Shrinkage and Selection Operator (“LASSO”), as done in Derenoncourt (2018). Below, we
report results obtained using this alternative procedure to construct the “push” version of the
instrument.
Results from equation (5) are reported in Table C.9. Columns 1 to 3 refer to, respectively,
the 1940-1950, the 1950-1960, and the 1960-1970 decade. All coefficients have the expected
sign. A higher share of land in cotton and of farms operated by tenants in 1940 are associated
with subsequent emigration. Somewhat surprisingly, however, the coefficient is not statistically
significant for the 1940-1950 decade, possibly because cotton mechanization was more prevalent
64Data on the cotton share comes from the Census of Agriculture, the vote share of Thurmond was takenfrom David Leip’s Atlas, while all remaining variables were collected from the County Databooks.
65Following Boustan (2016), in counties where the Black migration rate was above 100, we replace it withthe latter value. We also exclude counties with less than 30 Black residents in 1940. All results are robust toomitting these restrictions.
102
Table C.9. Zeroth Stage
Dep. Variable Net Black Migration Rate
(1) (2) (3)
Share Land in Cotton -0.012 -0.302∗∗ -0.163∗∗
(0.088) (0.123) (0.077)
Share Farms with Tenants 0.042 0.045 -0.173∗∗∗
(0.056) (0.064) (0.047)
WWII Spending per Capita 2.228∗∗∗ 0.393 0.046
(0.352) (0.359) (0.313)
Thurmond Vote Share -0.085∗∗ -0.083∗∗ -0.158∗∗∗
(0.037) (0.037) (0.042)
Share LF in Manufacturing -0.348∗∗∗ -0.248∗∗∗ -0.080
(0.090) (0.074) (0.070)
Share LF in Mining -0.440∗∗ -0.697∗∗∗ -0.522∗∗∗
(0.197) (0.179) (0.152)
Share LF in Agriculture -0.504∗∗∗ -0.486∗∗∗ -0.209∗∗∗
(0.050) (0.047) (0.045)
State Fixed Effects X X X
R-squared 0.256 0.283 0.163
Observations 1,163 1,163 1,163
Decade 1940-1950 1950-1960 1960-1970
Notes: The dependent variable is the net Black migration rate for southern counties for each decade indicatedat the bottom of the table. All regressors refer to 1940, except for Thurmond vote share, which is the vote shareof Thurmond in 1948 Presidential elections. All regressions include state fixed effects. See the appendix for thedefinition and source of variables included in the table. Robust standard errors, clustered at the county level,in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
in the 1950s (Grove and Heinicke, 2003). As in Boustan (2016), a higher share of the labor
force in mining and agriculture is associated with a larger emigration rate. Similarly, reflecting
a more hostile political environment, counties with a higher vote share for Thurmond in 1948
are predicted to have a higher emigration rate throughout the period. Finally, consistent with
WWII spending increasing labor demand, the Black in-migration rate is higher in counties with
more WWII contracts.
After estimating equation (5), we construct the predicted number of migrants by multiplying
the fitted values from (5) by the beginning of decade Black population. We then aggregate
these (predicted) flows to obtain the predicted number of Black migrants from each state in
each decade, Blsτ . Finally, we replace the actual number of Black migrants, Blsτ , with this
predicted value to construct a modified version of the shift-share instrument in equation (2) in
the main text.
C.4.2 Results
Table C.10 replicates our preferred specification for the Democratic vote share (columns 1-
2), turnout (columns 3-4), and CORE demonstrations (columns 5-6) using the push-factor
version of the instrument. In Panel A, we present 2SLS estimates, while in panel B we present
103
the associated first stage. Columns 1-3-5 report results obtained using the push instrument
constructed with the southern characteristics described above in the zeroth stage. Columns
2-4-6 turn to the version of the push instrument obtained by selecting predictors of southern
Black out-migration with the LASSO procedure (Derenoncourt, 2018).
Reassuringly, both versions of the instrument are strong, with the KP F-stat above con-
ventional levels. Moreover, the 2SLS estimates are in line with – in fact, for the Democratic
vote share and CORE demonstrations, stronger than – those presented in the main text. Also
in this case, Black in-migration has a positive and statistically significant effect on both the
Democratic vote share in Congressional elections and the probability of CORE demonstrations.
The effect on turnout remains positive, but small and imprecisely estimated. Taken together,
this exercise increases the confidence that our main results are not driven by local pull shocks
simultaneously correlated with the pre-1940 distribution of Black settlements across northern
counties.
Table C.10. Replicating Results with Push Instrument
Dep. variable Democratic Vote Share Turnout 1[CORE Demonstrations]
(1) (2) (3) (4) (5) (6)
Panel A. 2SLS
Change Black 1.856*** 2.298*** 0.420* 0.481* 0.050*** 0.056***
share (0.331) (0.504) (0.233) (0.276) (0.012) (0.013)
Panel B. First stage
Predicted Change 1.289*** 1.025*** 1.289*** 1.025*** 1.289*** 1.025***
Black Share (0.348) (0.303) (0.348) (0.303) (0.348) (0.303)
F-stat 13.68 11.40 13.68 11.40 13.68 11.40
Observations 3,418 3,418 3,418 3,418 3,418 3,418
Push Instrument Baseline LASSO Baseline LASSO Baseline LASSO
Notes: The table replicates the baseline specification using the version of the instrument constructed with
southern specific “push” factors. Columns 1-3-5 (resp. columns 2-4-6) report results for the “push” instrument
constructed using the baseline (resp. LASSO) procedure. The dependent variable is the change in Democratic
vote share, turnout, and probability of CORE demonstrations. Panel A reports 2SLS estimates, and Panel B
presents the first stage. All regressions are weighed by 1940 county population, and include interactions between
year dummies and: i) state dummies; ii) the 1940 Black share; and, iii) a dummy equal to 1 if the Democratic
vote share in 1940 was higher than the Republicans vote share. F-stat is the K-P F-stat for weak instruments.
Robust standard errors, clustered at the county level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05,∗ p< 0.1.
C.5 Robustness Checks on CD Results
C.5.1 Testing for Pre-Trends
Our main results in Table 7 showed that Black in-migration moved legislators’ ideology to
the left between 1940 and 1950. In Table C.11, we check that this pattern does not capture
104
pre-existing trends. Similar to what we did for the Democratic vote share and turnout in
Congressional elections (Section C.2 above), we construct the pre-period change in the ideology
scores, considering the first Congress after the New Deal, i.e. Congress 73. Then, we estimate
2SLS regressions for the pre-period change in the agnostic and the constrained version of the
scores against the instrumented change in the Black share, controlling for the same variables
included in our baseline specification (i.e. state dummies, and baseline: i) Black share; ii)
Democratic incumbency indicator; and, iii) ideology score).66
Table C.11. Testing for pre-trends: ideology scores
Dep. Variable Agnostic Scores Constrained Scores
(1) (2) (3) (4) (5) (6)
Panel A.2SLS
Change Black Share -0.307** -0.231* 0.037 -0.345*** -0.292* 0.035
(0.121) (0.134) (0.027) (0.130) (0.157) (0.033)
Panel B. First Stage
Predicted Change 1.006*** 0.985* 2.403*** 1.002*** 0.973* 2.416***
Black Share (0.379) (0.536) (0.275) (0.378) (0.535) (0.269)
F-stat 7.068 3.380 76.23 7.038 3.312 80.78
Observations 286 202 202 286 202 202
Specification Baseline Restricted Pre-Trends Baseline Restricted Pre-Trends
(1940-1960) (1940-1960)
Notes: Panel A reports 2SLS estimates for the change in agnostic (resp. constrained) ideology scores in columns
1 to 3 (resp. 4 to 6). Panel B reports first stage estimates. Columns 1 and 4 report the baseline specification
for Congress period 78-82 (see Table 7, columns 2 and 5), and columns 2 and 5 replicate this by restricting
attention to counties for which the change in the scores for the pre-period can be constructed. Columns 3 and
6 estimate 2SLS regressions for the change in the ideology scores between Congress 73 and Congress 78 against
the instrumented 1940-1960 change in the Black share. The pre-period is defined using the first Congress year
after the New Deal (i.e. Congress 73). All regressions are weighed by 1940 CD population, and include state
dummies, and the baseline: Black share, Democratic incumbency dummy, and ideology score. F-stat is the K-P
F-stat for weak instruments. Robust standard errors, clustered at the CD level, in parenthesis. Significance
levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05, ∗ p< 0.1.
To ease comparisons, we report the baseline specification for the 78-82 Congress period –
when Black in-migration induced a liberal shift in legislators’ ideology – in columns 1 and 4 for
the agnostic and the constrained version of the scores respectively. Since the pre-period change
in ideology could not be estimated for all CDs, in columns 2 and 5, we replicate columns 1
and 4 restricting attention to CDs for which the pre-trend check can be performed. When
doing so, the F-stat falls, suggesting that results should be interpreted with caution. However,
the point estimate remains negative, quantitatively close to that obtained for the full sample,
and statistically significant (with a p-value of 0.08 and 0.063 for agnostic and constrained
66As usual, regressions are weighed by baseline CD population.
105
scores respectively). Finally, in columns 3 and 6, we turn to the formal test for pre-trends.
Reassuringly, the point estimate is positive, close to zero, and imprecisely estimated. These
patterns indicate that the main results documented above are not influenced by a spurious
correlation between the instrument and potential pre-existing trends in ideology of legislators.
C.5.2 Alternative Timing Conventions for Congress Periods
In our baseline specification for the effects of the Great Migration on legislators’ ideology, we
mapped the 1940-1950 (resp. 1950-1960) change in the Black share to the 78-82 (resp. 82-88)
Congress period. This was done in order to include the longest periods without redistricting
while at the same time ending the analysis with the Congress that passed the CRA. We now
verify that our results above are robust to different timing conventions.
First, in Table C.12, we define the second period as ending with Congress 87 (rather than
Congress 88). The structure of the table mirrors that of Table 7 in the main text: columns 1 to
3 consider the agnostic version of the scores, while columns 4 to 6 focus on the constrained one.
Panel A reports 2SLS estimates, whereas Panel B presents the first stage. Not only results are
in line with those in the main text. But also, the point estimate for the stacked specification
doubles in size and becomes marginally significant. Second, in Table C.13, we define the end
of the first Congress period with Congress 83, in order to have two symmetric periods. While
the coefficient on the Great Migration remains highly negative and precisely estimated in the
first period, it becomes statistically significant (and positive) in the second period (reinforcing
the results on polarization discussed in Section 6.2.2 in the main text).
106
Table C.12. Changes in Legislators’ Ideology: Ending Analysis with Congress 87
Dep. Variable Change in Civil Rights Ideology (Lower values = more liberal ideology)
Agnostic Scores Constrained Scores
(Baseline mean: -0.872) (Baseline mean: -0.853)
(1) (2) (3) (4) (5) (6)
Panel A.2SLS
Change Black Share -0.094* -0.307** -0.007 -0.092* -0.345*** 0.011
(0.054) (0.121) (0.068) (0.054) (0.130) (0.069)
Panel B. First Stage
Predicted Change 1.473*** 1.006*** 1.811*** 1.456*** 1.002*** 1.785***
Black Share (0.441) (0.379) (0.556) (0.445) (0.378) (0.562)
F-stat 11.15 7.068 10.60 10.72 7.038 10.08
Observations 571 286 285 571 286 285
Congress Period 78-82; 82-87 78-82 82-87 78-82; 82-87 78-82 82-87
Notes: The dependent variable is the change in the civil rights ideology scores from Bateman et al. (2017)
– “Agnostic” scores in columns 1 to 3, and “Constrained” scores in columns 4 to 6. Lower (resp. higher)
values of the score refer to more liberal (resp. conservative) ideology (see also Bateman et al., 2017, for more
details). Columns 1 and 4 (resp. 2-3, and 5-6) estimate stacked first difference regressions (resp. first difference
regressions for Congress period 78-82 and 82-87). Panel A reports 2SLS results, while Panel B presents first stage
estimates. All regressions are weighed by 1940 congressional district population and control for state by year
fixed effects and include interactions between year dummies and: i) the 1940 black share in the congressional
district; ii) a dummy for Democratic incumbency in the 78th Congress in the district; and iii) the ideology score
in the district in the 78th Congress. First difference regressions do not include interactions with year dummies
since these are automatically dropped. F-stat refers to the K-P F-stat for weak instrument. Robust standard
errors, clustered at the congressional district level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ** p< 0.05,
* p< 0.1.
107
Table C.13. Changes in Legislators’ Ideology: Symmetric Congress Periods
Dep. Variable Change in Civil Rights Ideology (Lower values = more liberal ideology)
Agnostic Scores Constrained Scores
(Baseline mean: -0.872) (Baseline mean: -0.853)
(1) (2) (3) (4) (5) (6)
Panel A.2SLS
Change Black Share -0.041 -0.456*** 0.127** -0.044 -0.503*** 0.143***
(0.038) (0.165) (0.050) (0.040) (0.175) (0.054)
Panel B. First Stage
Predicted Change 1.471*** 1.006*** 1.808*** 1.454*** 1.002*** 1.782***
Black Share (0.440) (0.377) (0.554) (0.443) (0.377) (0.560))
F-stat 11.20 7.099 10.66 10.76 7.069 10.12
Observations 562 281 281 562 281 281
Congress Period 78-83; 83-88 78-83 83-88 78-83; 83-88 78-83 83-88
Notes: The dependent variable is the change in the civil rights ideology scores from Bateman et al. (2017)
– “Agnostic” scores in columns 1 to 3, and “Constrained” scores in columns 4 to 6. Lower (resp. higher)
values of the score refer to more liberal (resp. conservative) ideology (see also Bateman et al., 2017, for more
details). Columns 1 and 4 (resp. 2-3, and 5-6) estimate stacked first difference regressions (resp. first difference
regressions for Congress period 78-83 and 83-88). Panel A reports 2SLS results, while Panel B presents first stage
estimates. All regressions are weighed by 1940 congressional district population and control for state by year
fixed effects and include interactions between year dummies and: i) the 1940 black share in the congressional
district; ii) a dummy for Democratic incumbency in the 78th Congress in the district; and iii) the ideology score
in the district in the 78th Congress. First difference regressions do not include interactions with year dummies
since these are automatically dropped. F-stat refers to the K-P F-stat for weak instrument. Robust standard
errors, clustered at the congressional district level, in parentheses. Significance levels: ∗∗∗ p< 0.01, ** p< 0.05,
* p< 0.1.
C.5.3 Redistricting, Black Inflows, and Political Outcomes
One potential concern with results in Section 6 is that the decision of redistricting a CD may
have been at least partly driven by the arrival of African Americans. If this were to be the
case, and if redistricting had an effect on political outcomes, our results may be biased. As
noted in Appendix B, until 1964 (i.e. the end of our sample period), redistricting was unlikely
to be strategic (Engstrom, 2013), and was typically mandated at the state level. We exploit
the fact that between Congress 78 and Congress 82, five states in our sample (Arizona, Illinois,
New York, Maryland, and Pennsylvania) required their CDs to redistrict, and test whether
redistricting was systematically correlated with either Black inflows or changes in political
conditions (e.g. party switches, changes in legislators’ ideology, etc.).67
In Table C.14, the dependent variable is a dummy equal to 1 if a CD belongs to a state
67This check cannot be performed between Congress 83 and Congress 88 because most CDs were subject toredistricting in this period.
108
that did not mandate redistricting, and is regressed against: i) changes in the Black share
(with OLS in column 1 and with 2SLS in column 2); ii) a dummy if the CD underwent a party
switch; iii) the change in the Bateman et al. (2017) ideology score (column 4); and iv) the
number of discharge petitions signed per legislator (column 5). Since the dependent variable
varies at the state level, we cannot control for state fixed effects; yet, we include (as in our
baseline specifications) the 1940 Black share and the 1940 Democratic dummy. Reassuringly,
the coefficient is never statistically significant, does not display any systematic pattern, and
is always quantitatively small. Overall, this exercise thus suggests that neither changes in
the Black share nor changes in political conditions were systematically associated with state-
mandated redistricting.
Next, we inspect more directly the possibility that Black inflows led to strategic gerry-
mandering across CDs. In particular, we rely on the measure of (non-)compactness recently
introduced by Kaufman et al. (2017), which is based on the geographic shape of CDs, and
captures the “compactness evaluations” made by judges and public officials responsible for re-
districting.68 We prefer to use this measure, instead of an alternative proxy based on the vote
distribution, because it provides evidence of (potential) gerrymandering at the CD level. In
turn, this allows us to investigate the relationship between non-compactness and Black inflows.
The measure of compactness can take values between 1 and 100, with higher values indicating
less compact districts, i.e. a higher probability of gerrymandering.
We start by analyzing descriptively the trends of non-compactness between Congress 71 and
Congress 90 in Figure C.4. Consistent with the existing literature discussed in Appendix B,
for the period considered in our analysis – between Congress 78 and Congress 88 – average
compactness changes very little. Reassuringly, other aggregate measures, such as the standard
deviation and the interquantile range, do not show any detectable changes either (not shown).
Interestingly, and again consistent with existing studies, non-compactness starts to increase
precisely after Congress 88, suggesting that after 1964 strategic gerrymandering might have
become gradually more common.
Then, we study the relationship between Black inflows and non-compactness during our
sample period. To do so, we proceed as follows. First, we assign the 1940-1950 (resp. 1950-
1960) change in the Black share to each Congress in the 78-82 (resp. 83-88) Congress period.
Second, we estimate 2SLS regressions where the dependent variable is the measure of non-
compactness specific to each Congress number (for the relevant decade) and the main regressor
of interest is the instrumented change in the Black share. Figure C.5 reports the implied
2SLS coefficients (with corresponding 95% confidence intervals) from previous regressions cor-
responding to a one standard deviation change in the Black share.
If the arrival of African Americans induced northern politicians to strategically change
the boundaries of CDs, we would expect the association between changes in the Black share
and non-compactness to increases over time. Reassuringly, there is no statistically significant
relationship between the change in the Black share and the measure of non-compactness in
any Congress year. Our estimates are also quantitatively small. For instance, one standard
deviation increase in the Black share (around 2.8 percentage points) increases compactness of
68We thank the authors for making their codes available to us.
109
Congress 78 by 2 points – a negligible effect when compared to a mean of 45 and to a standard
deviation of 16. Moreover, coefficients do not display any increasing trend over time, suggesting
that strategic gerrymandering in response to Black arrivals was very unlikely to occur during
our sample period.
Table C.14. Redistricting Checks
Dep. Variable 1[Non-Redistricting State]
(1) (2) (3) (4) (5)
Change Black Share 0.014 0.039
(0.013) (0.038)
Party Switch 0.084
(0.061)
Change Ideology Scores -0.007
(0.049)
Signatures on Discharge Petitions -0.035
(0.023)
F-stat 17.31
Observations 286 286 286 286 298
Notes: The dependent variable is a dummy equal to 1 if the CD belongs to a state that did not mandate
redistricting between Congress 78 and Congress 82. In columns 1 and 2, the main regressor of interest is the
change in the Black share during the 1940-1950 decade. Column 1 (resp. column 2) presents OLS (resp. 2SLS)
results. Columns 3, 4, and 5 regress the redistricting state dummy against, respectively, a dummy equal to 1 if
the CD experienced a party transition during the 78-82 Congress period, the change in Bateman et al. (2017)
scores, and the signatures on discharge petitions per legislator. All regressions control for the 1940 Black share,
and for a dummy equal to 1 if the Democratic vote share in 1940 was higher than the Republicans vote share.
Robust standard errors, clustered at the CD level, in parenthesis. Significance levels: ∗∗∗ p< 0.01, ∗∗ p< 0.05,∗ p< 0.1.
110
Figure C.4. Average Non-compactness, Congress 71st-90th
Notes: The figure presents the area-weighted average non-compactness for each
Congress between Congress 71 and Congress 90. The red, vertical lines, correspond-
ing to Congresses 78 and 88 isolate the sample period considered in our paper.
111
Figure C.5. Black In-Migration and Non-compactness
Notes: The figure presents the 2SLS coefficient with the corresponding 95% confi-
dence interval implied by one standard deviation change in the Black share during
the corresponding decade. The dependent variable is the CD non-compactness score
from Kaufman et al. (2017). The main regressor of interest is the 1940 to 1950 (resp.
1950 to 1960) change in the Black share for Congresses between 78 and 82 (resp.
between 83 and 88), and is instrumented with the shift-share instrument described
in the text. All regressions control for state dummies, for the 1940 Black share,
and for a dummy equal to 1 if the district was represented by a Democrat in each
Congress.
112
D Appendix – Survey Data
D.1 The American National Election Studies (ANES)
The American National Election Studies (ANES) is a cross-sectional, nationally representative
survey conducted since 1948 by the University of Michigan every two or four years depending
on the waves. As noted in Gentzkow (2016), the ANES is considered the “gold standard”
when it comes to measure political ideology and cultural or social attitudes of Americans in
the second part of the twentieth century. The ANES asks questions on demographics, party
affiliation, political attitudes, and ideology. Moreover, and crucially for our purposes, since the
mid-late 1950s, respondents are asked about their views on civil rights legislation and racial
equality and, in some instances, about their attitudes towards integration.69
In each wave, between 1,500 and 2,000 respondents were interviewed. Restricting the
sample to whites living in non-southern states leaves us with an average of roughly 1,000
individuals for whom we can consistently include the following controls: marital status, gender,
and fixed effects for education and age.70 In principle, additional characteristics, such as
union status, party affiliation and identification, and state of birth, are available. Since these
may be endogenous to Black migration, however, we do not control for them in our baseline
specification. Most of our analysis uses data from the surveys of 1960 and 1964, but, in a
few cases, we were able to obtain data also from other years. Before 1978, the ANES never
reported the county, but only the state, of residence of respondents. For this reason, our
analysis is conducted at the state level.
Table D.1 presents the questions considered to measure racial attitudes and views towards
civil rights. The first column presents the name of the variable; the second one includes the
exact wording of the question; and the last column lists the years for which the question was
available. The first variable listed, “Most Important Problem” refers to an open-ended ques-
tion in which respondents were asked what they considered (up to) the three most important
problems for the US in the year of the survey. From such open-ended question the ANES
created one specific category that includes racial and public order related issues. For 1960 and
1964, the ANES coded respondents’ answers in categories that reflected their attitudes towards
civil rights and integration.71 We use the ANES pre-classified category “Pro integration - anti
discrimination in schools, employment, etc.” to create a dummy equal to one if the respon-
dent believes that promoting integration in schools, employment, etc. is one of the top three
problems facing the country in that survey year. This is the variable 1[Pro Civil Rights: Most
Important Problem] considered in Table 5 in the main text.
Table 5 verifies that the variable 1[Pro Civil Rights: Most Important Problem] is negatively
correlated with opposition to school and housing or work integration. Again for 1960 and
1964, from the ANES survey, we created dummies (reported in the second and third row of
69More details on ANES sampling methodology and data are available at http://www.electionstudies.org/wp-content/uploads/2018/04/nes012492.pdf
70We create dummies for: high school dropouts; high school graduates; having at least some college; havingat least a college degree.
71Unfortunately, for other years, it was not possible to tell whether the respondent identified civil rights assomething to promote or instead as an issue that was undesirable to her.
113
Table D.1) equal to one if the respondent, respectively, agreed that the federal government
should not intervene to promote racial integration in schools and disagreed with the idea that
the government should promote racial integration in housing and labor markets.
As discussed in the main text, we exploit ANES questions concerning political preferences
in surveys in years between 1956 and 1964. In particular, individuals were asked to indicate
the party they had voted (resp. intended to vote) in the previous (resp. upcoming) elections.
From this variable, we create a dummy equal to one if respondents answered that they voted
or intended to vote for the Democratic Party (Vote Democratic). Similarly, the ANES asked
respondents whether they identified with either the Democratic or the Republican Party. As
for voting behavior, we create a dummy equal to one if respondents identified themselves with
the Democratic Party (Identify Democratic). We use these outcomes in the analysis reported
in Table A.19.
Finally, for 1964 only, ANES respondents were asked about their feeling thermometers
towards different political and socio-demographic groups, including the Democratic Party, labor
unions, Blacks and the NAACP. Thermometer values are such that higher values refer to
warmer feelings towards members of the group.72 We use the answers given by respondents in
Appendix Table A.24.
D.2 Gallup
We validate the results obtained using the ANES with Gallup, which elicited respondents’
views about salient political and social issues since 1935.73 As for the ANES, also Gallup is
a repeated cross-sectional dataset from which individual level characteristics are available.74
Starting from the mid-1950s, Gallup asked questions about racial attitudes. As discussed
extensively in Kuziemko and Washington (2018), Gallup data have been only recently made
available due to the efforts of the Roper Center, which digitized hundreds of historical surveys.75
As for the ANES, we restricted attention to white respondents living in non-southern states in
years before 1965. In practice, so as to keep the sample consistent across questions, we focused
on years 1963 and 1964, when different questions, comparable to those from the ANES, were
asked. We report the wording and the survey years for which these two questions are available
in Panel B of Table D.1.
Starting from the top of Panel B, Gallup respondents who had at least one child in school
were asked whether they would object to send their kids to a school with few, half, and more
than half Black pupils. Parents who responded that they would not object to sending their
kids to a school with few Black students were subsequently asked if they would object to a
situation with half Black pupils in the school. If they had no objections to such question, they
were asked about a situation in which the school was more than half Black. Most parents
72The ANES asked respondents about their feeling thermometers towards the two parties also in years otherthan 1964. However, since we are interested in studying whites’ racial attitudes, we limit our analysis to 1964,i.e. the only year for which both political and racial groups or organizations were included.
73See also https://ropercenter.cornell.edu/featured-collections/gallup-data-collection.74With the exception of union membership, marital status, and state of birth, all individual characteristics
available in the ANES (see Appendix D.1) are available for Gallup as well.75More information about Roper Center data can be found here: https://ropercenter.cornell.edu/. We
thank Kathleen Joyce Weldon for invaluable help in the data collection and data cleaning process.
114
(90%) did not object to send their kids to schools with only a few Black pupils. Instead, more
heterogeneity existed when parents were asked about a situation in which half or more than
half of the school were racially mixed. Specifically, 30% of parents who did not object to send a
kid to a school that had few Black pupils were against sending their kid to a school where half
of the pupils were Black. Of those that did not object to send their kid to a school where half
of the pupils were Black 38% were against a situation in which more than half of the pupils in
the school were Black.
Given these patterns, we decided to focus on the answer to the scenario in which half
of the pupils in the school were Black. In our view, and consistent with existing evidence
(Sugrue, 2008, 2014), racial mixing was not perceived as a threat when (school or neighborhood)
integration entailed only a limited number of Blacks. Instead, racial animosity and whites’
backlash was more likely to emerge as the share of Blacks in the local (white) community
increased. The variable 1[Object to Half Black Pupils in School] at the top of column 1 in
Table A.20 is thus a dummy equal to 1 if parents did object to sending their kids to a school
with at least half of students being Black.76
The second question used in our analysis is meant to capture whites respondents’ acceptance
of racial diversity in politics. Specifically, as in Kuziemko and Washington (2018), we consider
whether respondents would vote for a Black candidate had their party nominated the individual
for the Presidential race (see second row in Panel B of Table D.1).77 In column 2 of Table A.20,
we create a dummy equal to one if respondents answered that they would vote for a Black
candidate, 1[Vote for Black Candidate].78
In 1964, given the prominence of the issue, Gallup questionnaires included a question about
the Civil Rights Act (CRA). Among the about 1,000 respondents, approximately 70% of them
did approve the law just passed by Congress. We create a dummy equal to one if a respondent
supported the CRA (Approve Civil Rights Act in Panel B of Table D.1). This variable is
considered as outcome in column 3 of Table A.20.
Finally, we consider a question that elicits respondents’ view on how the Kennedy Admin-
istration was handling the process of racial integration. Specifically, we create a dummy equal
to one if an individual stated that in her view, racial integration was proceeding “at the right
pace or not fast enough” (see the last row in Table D.1, Panel B). This variable is used as
outcome in column 4 of Table A.20.79
76The sample size is relatively small – 851 respondents – since only parents with kids in school were askedthis question.
77In 1963 this question was asked to around 2,000 respondents.78Kuziemko and Washington (2018) investigate whites’ respondents to this question also for years after 1965.
Instead, in order not to confound our results with potential whites’ backlash we stop in 1963 – the last yearbefore the passage of the CRA.
79As it appears from Table A.20, this question is available for a significantly larger number of respondents(more than 17,000) relative to all other questions. This is because the question was asked repeatedly in 1963.
115
Table D.1. Questions from Survey Data
Variable Name Wording Years
Panel A. ANES
Most Important Problem What would you personally feel are the most important problems the government should tryto take care of when the new president and congress take office in January. (Do you think ofany other problems important to you)
1960 and1964
Against School Integration The government in Washington should stay out of the question of whether white and [Black]children go to the same school.
1960 and1964
Against Work and ResidentialIntegration
If [Blacks] are not getting fair treatment in jobs and housing, the government should see to itthat they do.
1960 and1964
Vote Democratic 1 if voted/intend to vote for the Democratic Party in the last/upcoming Presidential Elections 1956-1964
Party Identification Was there ever a time when you though of yourself as a Democrat or a Republican? 1956-1964
Feeling Thermometer Towards[Group]
There are many groups in America that try to get the government of the American people tosee things more their way. We would like to get your feelings toward some of these groups...Where would you put (group) on the thermometer?
1964
Panel B. Gallup
Object to Half Black Pupils inSchool
Would you, yourself, have any objection to sending your children to a school where half of thechildren are [Black]
1963
Black Candidate There’s always much discussion about the qualifications of presidential candidates - theireducation, age, religion, race and the like... If your party nominated a generally well-qualifiedman for president and he happened to be a [Black] would you vote for him
1963
Approve Civil Rights Act As you know, a civil rights law was recently passed by Congress and signed by the President.In general, do you approve or disapprove this law?
1964
Racial Integration at the RightPace/Not Fast Enough
Do you think the Kennedy Administration is pushing racial integration too fast or not fastenough?
1963
Notes: Panel A (resp. B) lists variables and questions taken from the ANES (Gallup). The wording reported for variables Most Important Problem, Against SchoolIntegration, and Against Work and Residential Integration in Panel A is taken from the 1960 survey, but remains almost identical in all other years considered.
116